”mu

.. ,— 2-4.. - 41.»:

3—,

 

_
.-
23.... .

 

‘: 2;:

,
4:“;
‘.’

'O

331-"

u» -
up».

d ..

"3‘1

u
.0

.
dlv

w
'3‘.
;..

...- J.
u—o‘e‘

513‘ '
$5.57
hfl

(-J‘

     

‘r
' §
.' .: --~« 7

M ’33: .

-,--
4,»:

a V
fiégh‘g ‘

1
o

a .
rum. ’ a"; b 5’4“
“ “21%?2é' :7
a
r .

fifty

 

.1
u

u

. ‘ Rf
if;
JEN»;

.

THESIS
2

lllllllllllllllWilliwill

3 01688 4664

This is to certify that the

thesis entitled

XANTHOMONAS PATHOVARS IDENTIFICATION THROUGH A
NEURAL NETWORK-BASED GENOMIC FINGERPRINT
CLASSIFICATION SYSTEM

presented by

Fei Ni Tuang

has been accepted towards fulfillment
of the requirements for

 

 

M.S degree in Biosystems Engineering

Major professd

Date W Zé/ /476>

 

0-7639 MS U is an Affirmative Action/Equal Opportunity Institution

 

 

LIBRARY

Michigan State
Unlverslty

 

 

 

PLACE IN RETURN BOX
to remove this checkout from your record.
TO AVOID FINES return on or before date due.

 

DATE DUE DATE DUE DATE DUE

 

 

 

MAR 2 3. 20W

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1/” WWW“

XANTI-IOMONAS PATHOVARS IDENTIFICATION THROUGH A
NEURAL NETWORK-BASED GENOMIC FINGERPRINT
CLASSIFICATION SYSTEM

By

Fei Ni Tuang

ATHESIS

Submitted to
Michigan State University
In partial fulfillment of the requirements
for the degree of

MASTER OF SCIENCE

Department of Agricultural Engineering

1998

ABSTRACT
XANT H OMONAS PATHOVARS IDENTIFICATION THROUGH

A NEURAL NETWORK-BASED GENOMIC FINGERPRINT
CLASSIFICATION SYSTEM

By

Pei Ni Tuang

A genomic fingerprint classification system was developed to identify 63
Xanthomonas pathovars. Three sets of genomic fingerprints generated from repetitive
DNA sequence-based polymerase chain reaction (rep-PCR), using BOX, ERIC and REP
primers, were used in this research. In addition, a fourth set of BER fingerprints was
formed by linearly combining the BOX, ERIC and REP fingerprints. Mean and wavelet
filter techniques were used to reduce noise on the fingerprints. Several backpropagation
neural network (BPN) classifiers were trained using the BOX, ERIC, REP and BER
original fingerprints and filtered fingerprints. Both mean and wavelet filtering helped
improve the recognition rates. Wavelet filtering was better at reducing misclassification
error rates, and mean filtering was better at reducing false rejection error rates. The
average top-2 recognition rates of BOX, ERIC, REP and BER BPN classifiers were 95%,
‘93%, 92% and 98%, respectively. By combining the results of the BOX, ERIC and REP
BPN classifiers with the lowest misclassification error rates, a top-1 recognition rate of
95% was achieved together with a misclassification error rate of 0.57% and a false

rejection rate of 4.3%.

Copyright by
FBI NI TUAN G
1998

ACKNOWLEDGEMENTS

I would like to express my sincere gratitude to my committee members: Dr.
Evangelyn Alocilja (Biosystems Engineering), Dr. John Gerrish (Biosystems
Engineering), and Dr. John Weng (Computer Science) for their insight, guidance and
support in the course of this study.

My sincere thanks to Jan Rademaker and Dr. Frans de Bruijn of MSU DOE-Plant
Research Laboratory for providing the genomic fingerprint data and giving me the
opportunity to learn the rep—PCR genomic fingerprinting technique in the laboratory.

I am also appreciative for the funding provided by Dr. Robert von Bemuth,
through the Manure Management and Nutrient Balance for Dairy project, and the
fellowship provided by the College of Agriculture and Natural Resources.

I would like to thank the faculty, staff and students of the Department of
Agricultural Engineering for providing a challenging learning environment. Special
thanks to my friends at MSU: Thomas Moen, Tse-chia Yu, Erica Yang, Ismail Kavdir,
Pankaj Jagtap, Jim Schaper and Takako Inagaki for their support and encouragement.

Finally, a very special thank you to my parents and my entire family for their

never-ending guidance, support, encouragement and love.

iv

TABLE OF CONTENTS

LIST OF TABLES .......................................................................................................... vii

LIST OF FIGURES ........................................................................................................ viii

LIST OF ABBREVIATIONS .......................................................................................... xi

INTRODUCTION ............................................................................................................. 1

Objectives .............................................................................................................. 3
CHAPTER I

LITERATURE REVIEW .................................................................................................. 4

1.1 Bacterial Taxonomy and Classification System .............................................. 4

1.2 The genus Xanthomonas ................................................................................. 7

1.3 rep-PCR Genomic Fingerprinting ................................................................... 8

1.4 Digital Image Processing ............................................................................... 10

1.5 Backpropagation Neural Network ................................................................. 14

1.6 Performance Evaluation ................................................................................ 18

1.7 Systems Analysis ........................................................................................... 19
CHAPTER H

MATERIALS AND METHODS ..................................................................................... 22

2.1 Data Collection .............................................................................................. 22

2.2 Systems Analysis ........................................................................................... 24

2.2.1 Data Modeling ....................................................................................... 25

2.2.2 Process Modeling ................................................................................. 29

2.3 Data Processing .......................................................................................... 31

2.3.1 Data Input ......................................................................................... 31
2.3.2 Image Filtering .................................................................................. 32
2.4 BPN Classifier Design ................................................................................ 33
2.4.1 Data Partitioning ................................................................................ 33
2.4.2 BPN Training .................................................................................... 35
2.5 BPN Classifier Testing ............................................................................... 37
2.6 Performance Evaluation ............................................................................. 38
2.7 Graphical User Interface ............................................................................. 39
CHAPTER III
RESULTS AND DISCUSSION ................................................................................... 41
3.1 Image Filtering ........................................................................................... 41
3.2 BPN Training ............................................................................................. 46
3.3 BPN Classifier Testing ............................................................................... 48
3.4 Perfonmnce Evaluation ............................................................................. 50
3.5 User Scenario ............................................................................................. 59
CHAPTER IV
SUMMARY AND CONCLUSIONS ........................................................................... 68
CHAPTER V
RECOMMENDATIONS ............................................................................................. 7O
APPENDICES ............................................................................................................ 71
Appendix A. rep-PCR Genomic Fingerprint Data ............................................ 72
Appendix B. Database and Data Files .............................................................. 86
Appendix C. Program Listing ........................................................................... 93
LIST OF REFERENCES ........................................................................................... 104

vi

LIST OF TABLES

Table 1.1 Data Modeling Concepts and Definitions ..................................................... 21
Table 2.1 Notations of Data Flow Diagram .................................................................. 29
Table 3.1 BPN Training Results (Set 1) ....................................................................... 46
Table 3.2 BPN Training Results (Set 2) ....................................................................... 47
Table 3.3 BPN Classifiers Performance Evaluation ..................................................... 51
Table 3.4 Performance of Combined BPN Classifiers .................................................. 52
Table 3.5 Bacterial Strains Wrongly Identified or Rejected by BEROO-l BPN ............. 53
Table 3.6 Bacterial Strains Wrongly Identified or Rejected by DAT02B-l BPN .......... 54
Table 3.7 Bacterial Strains Wrongly Identified or Rejected by DATOBE-l BPN ........... 55
Table 3.8 Bacterial Strains Wrongly Identified or Rejected by DAT02R-1 BPN .......... 56
Table 3.9 Bacterial Strains Wrongly Identified or Rejected by

Combined BOX, ERIC and REP BPN Classifiers ........................................ 57
Table 3.10 Speed Performance of rep-PCR Genomic Fingerprint

Classification System .................................................................................. 58
Table A] List of rep-PCR Genomic Fingerprint Data ................................................. 72
Table B.1 rep-PCR Genomic Fingerprint Database Properties ..................................... 86
Table B.2 BER Fingerprint Database Properties .......................................................... 88

vii

LIST OF FIGURES

Figure 1.1 Wavelet Multilevel Decomposition Tree ....................................................... 12
Figure 1.2 Model of a Neuron ......................................................................................... 14
Figure 1.3 Structure of a BPN ......................................................................................... 15
Figure 1.4 Systems Development Life Cycle .................................................................. 19
Figure 2.1 BOX-PCR genomic fingerprint (LMG 12141) .............................................. 23
Figure 2.2 BER Fingerprint (LMG 12141) ..................................................................... 23
Figure 2.3 Entity Relationship Diagram of rep-PCR Fingerprints .................................. 27
Figure 2.4 Entity Relationship Diagram of BER Fingerprints ........................................ 28
Figure 2.5 DFD of rep-PCR Genomic Fingerprint Classification System ...................... 30
Figure 2.6 DFD of Data Input ......................................................................................... 31
Figure 2.7 Scaling Function and Wavelet Function of db8 ............................................. 32
Figure 2.8 DFD of Image Filtering ................................................................................. 33
Figure 2.9 DFD of Data Partitioning ............................................................................... 34
Figure 2.10 The Architecture of a BPN Classifier .......................................................... 35
Figure 2.11 DFD of BPN Classifier Testing ................................................................... 37

Figure 2.12 A Tree Representation of the BPN Classification Preformance Analysis 39

Figure 3.1 BOX Fingerprint (LMG 12141) Filtered with Mean Filter ........................... 41
Figure 3.2 ERIC Fingerprint (LMG 12141) Filtered with Mean Filter ........................... 42
Figure 3.3 REP Fingerprint (LMG 12141) Filtered with Mean Filter ............................ 42

viii

Figure 3.4 Reconstructed Image, Approximation and Details before Thresholding
(BOX Fingerprint — LMG 12141 with db8 Wavelet at Level 4
Decomposition) .......................................................................................... 43

Figure 3.5 Detail Coefficients and The Associated Threshold for Noise Reduction
(BOX Fingerprint - LMG 12141 with db8 Wavelet at Level 4

Decomposition) .......................................................................................... 43
Figure 3.6 BOX Fingerprint (LMG 12141) Filtered with db8 Wavelet at
Level 3 Decomposition .............................................................................. 44
Figure 3.7 ERIC Fingerprint (LMG 12141) Filtered with db8 Wavelet at
Level 3 Decomposition .............................................................................. 44
Figure 3.8 REP Fingerprint (LMG 12141) Filtered with db8 Wavelet at
Level 3 Decomposition .............................................................................. 44
Figure 3. 9 BOX Fingerprint (LMG 12141) Filtered with db8 Wavelet at
Level 4 Decomposition .............................................................................. 45
Figure 3.10 ERIC Fingerprint (LMG 12141) Filtered with db8 Wavelet at
Level 4 Decomposition ............................................................................. 45
Figure 3.11 REP Fingerprint (LMG 12141) Filtered with db8 Wavelet at
Level 4 Decomposition ............................................................................. 45
Figure 3.12 BER Fingerprint (LMG 12141) Filtered with db8 Wavelet at
Level 3 Decomposition ............................................................................. 46
Figure 3.13 DATOOB BPN Training and Generalization Sum Square Errors ................ 48
Figure 3.14 BPN Output Scores of a Bacterial Strain with an Unique Identification
(DATOOB-l — LMG 471) .......................................................................... 49
Figure 3.15 BPN Output Scores of a Bacterial Strain with No Identification
(DATOOB-1 — LMG 471) .......................................................................... 50
Figure 3.16 BPN Output Scores of a Bacterial Strain with Two Identification
(DATOOB-l - LMG 471) .......................................................................... 50
Figure 3.17 A Snapshot of rep-PCR Genomic Fingerprint Classification System ......... 59
Figure 3.18 Data Input Screen ..................................................................................... 60
Figure 3.19 Data Selection Screen ............................................................................... 61

ix

Figure 3.20 Image Filtering Settings Screen ................................................................. 61

Figure 3.21 Filtering Output Screen ............................................................................. 62
Figure 3.22 Target Pathovar Screen ............................................................................. 63
Figure 3.23 Training / Testing Sets Screen ................................................................... 63
Figure 3.24 Non-target Pathovar Screen ...................................................................... 64
Figure 3.25 BPN Settings Screen ................................................................................. 64
Figure 3.26 Classification Setting Screen ..................................................................... 65
Figure 3.27 BPN Output Screen ................................................................................... 66
Figure 3.28 BPN Classification Analysis Screen .......................................................... 67
Figure 3.29 Classification Comparison Screen ............................................................. 67
Figure A] Sample rep-PCR Genomic Fingerprint Data File (First Two Strains) .......... 84

LIST OF ABBREVIATIONS

DNA .................................................................................... Deoxyribonucleic acid
PCR ................................................................................ Polymerase chain reaction
rep-PCR ...................................................... Repetitive DNA sequence-based PCR
BPN ..................................................................... Backpropagation neural network
GUI .................................................................................... Graphical user interface
Pathovar ..................................................................................... Pathogenic variant
bp ............................................................................................................. Base pairs
IE ...................................................................................... Information Engineering
ERD ............................................................................. Entity relationship diagram
DFD ........................................................................................... Data flow diagram
SSE ............................................................................................... Sum square error
SQL ............................................................................. Structured Query Language

xi

INTRODUCTION

Bacteria are one of the smallest life forms on Earth, but they have significant
impacts on large organisms and on the ecosystem as a whole. Some bacteria, such as the
Xanthomonas Species, are plant pathogens. They can cause diseases in a wide variety of
agriculturally and economically important plants. Xanthomonas infections of fruits and
vegetables cause considerable loss in yield (Hayward, 1993). Identification of the
bacterial species and pathogenic variants (pathovars) will lead to disease control, and
prevention of losses in crop yield. In addition to its crucial role in pathological studies,
bacterial identification also plays an important role in ecological studies, biodiversity
assessment and microbial screening for industrial application.

Among the latest developments in bacterial taxonomy are: (1) molecular
identification techniques based on genotypic features and (2) computer-assisted
classification with numerical analysis (Vauterin et al., 1993). Repetitive deoxyribonucleic
acid (DNA) sequence-based polymerase chain reaction (rep-PCR) genomic fingerprinting
technique is capable of rapidly generating large quantities of highly discriminated DNA
fingerprints. The fingerprinting method, coupled with computer-assisted cluster analysis,
was found to successfully classify and identify Xanthomonas pathovars (Rademaker et
al., 1997). In cluster analysis, a Similarity matrix of the different fingerprints is used to
generate a dendrogram. In the dendrogram, strains of bacteria belonging to the same

pathovar are clustered together. A new bacterial strain is identified by comparing its

fingerprint to fingerprints of other pathovars in a reference library. The bacterial strain is
assigned to the pathovar with the closest match based on the dendro gram (Rademaker
and de Bruijn, 1997).

Backpropagation neural network (BPN) is a type of artificial neural network that
is suitable for use as a pattern classifier. The BPN is a feedforward network made up of a
collection of interconnected nodes or processing elements (Castleman, 1996). It can be
trained to recognize and classify complex patterns. Using a supervised training approach,
training data consisting of input and desired output pairs are presented to the network
repetitively, and the connection weights are adjusted until the error between the network
output and desired output converges (Haykin, 1994). A network can be used to recognize
patterns for which it is trained.

The use of rep-PCR genomic fingerprints with BPN can be an efficient approach
to identify newly isolated bacterial strains. A trained network can identify bacterial
pathovars based on its internal representation, without comparing them to other
fingerprints in the reference library. The massive computation needed to learn the
classifications is done only once during the training phase. The advantage of using the
BPN approach over the cluster analysis approach will be appreciated when the size of the
reference library and the number of strains that needs to be identified are large.

Noise and variability in the genomic fingerprints are factors affecting the
accuracy of BPN classification. Noise is introduced in the process of fingerprinting, and
is caused by some physical, chemical and biological factors in PCR reactions, gel

electrophoresis and photography. Filtering techniques such as local averaging mean filter

and wavelet analysis offer possible noise reduction. Classification accuracy may be
improved with the use of these filtering techniques.

Since the amount of data to be organized and processed for BPN training, testing,
and classification, is large, an information system is essential. Graphical visualization
capability is also necessary for visualizing fingerprint patterns and examining within
class variations. In addition to a classifier component, a good classification system should
have an integrated database, data processing and data visualization components, and a
graphical user interface (GUI) that links every component together and serves as a means
for the user to communicate with the system. This will lead to a more efficient and

standardized classification procedure since every component is built into the system.

Objectives
The objectives of this research are:
1. To build a rep-PCR genomic fingerprint classification system that is rapid,
accurate and easy to use.
2. To investigate the effect of noise reduction on classification accuracy.
3. To evaluate the performance of the several BPN classifiers in identifying

Xanthomonas pathovars.

Chapter 1

LITERATURE REVIEW

1.1 Bacterial Taxonomy and Classification Systems

Taxonomy is the Study of the classification of living things (King and Stanfield,
1997). It involves the study of relationships among living things and their arrangement
into categories. Bacterial classification is important for several reasons. Firstly, it serves
as a means of summarizing and cataloguing information about an organism. Secondly, it
forms the basis for the recognition of new isolates. Thirdly, it provides an insight into the
origin and evolutionary pathways of bacteria and higher organisms (Priest and Austin,
1993). Fourthly, it is fundamental in understanding ecological processes in a biosphere
with diverse species and complex biological interactions. Fifthly, it plays an important
role in disease control and microbial screening for industrial applications (T owner and
Cockayne, 1993).

To effectively meet the classification objectives, a classification method must
satisfy the following three criteria: high stability, good predictivity and high objectivity
(Priest and Austin, 1993). While biological classifications are subject to change in time
(this is unavoidable due to the discovery of Species or the collection of new data), there
are still some measures that can be taken to enhance classification stability. Firstly, all

relevant objects should be included before trying to form groups. Secondly, the

significant characteristics of the objects and all their variations should be included in the
study (Pankhurst, 1991). To enhance predictivity and to include more generalizations
about the taxa involved, classifications should be based on high information content. To
increase objectivity, the development of a classification Should be empirical, reproducible
and scientifically based (Priest and Austin, 1993). While the development of a
classification system is often data dependent, these criteria and measures provide some
general guidelines.

The development of bacterial taxonomy can be traced back to the seventeenth
century in the work by Morison in 1672 and Ray in 1688 (Pankhurst, 1991). The early
classifications were based on morphological information as obtained from light
microscopy. Following the interest in morphology, pure culture techniques were
developed to study bacterial metabolism and physiological information (Priest and
Austin, 1993). The traditional bacterial classifications were mainly based on phenotypic
properties such as biochemical, nutritional and physiological features. In the 19603 and
early 19705, numerical taxonomy was developed to provide more objective analysis and
to automate the process (Goodfellow et al., 1997). Some computer-assisted numerical
classification methods include multi-access and hypertext key, expert systems (Edwards
and Morse, 1995), matching and clustering using similarity coefficient, maximum
likelihood classification, discriminant analysis (Pankhurst, 1991) and neural networks
(Weeks and Gaston, 1995). With computer-assisted classification, a comparison of large
numbers of features and strains is allowed (Vauterin etal., 1993). The numerical methods

greatly enhance classification accuracy and efficiency.

In a survey paper by Wilkins et al. (1996), two statistical methods, including
parametric Gaussian multivariate statistical method (GAUSS) and non-parametric K-
nearest neighbour method (KNN), and four artificial neural network paradigms, including
multilayer perceptron networks (MLP), learning vector quantization networks (LVQ),
radial basis function network (RBF) and asymmetric radial basis function network
(ARBF), were compared for their ability to identify phytoplankton species from flow
cytometry data. From the results of the studies, all classifiers had similar identification
success, with recognition rate within 3.4% of each other. The relative merits of each
classifier were compared. The MLP was found to be reliable, robust, sirnple-to-use and
rapid in operation, but had a longer training time in comparison with other methods.
Training is not required for the KNN classifier. However, it is slow in operation and
requires storage of a large representative training set (Wilkins et al., 1996). The MLP is
the most popular class of multilayer feed-forward network, which is also called
backpropagation neural network (BPN) after its back-propagation learning algorithm
(Jain et al., 1996). .

Recent advances in molecular biology make possible the classification of bacteria
using detailed genotypic properties. Some nucleic acid methods, such as DN A-DN A
hybridization, DNA-RNA hybridization and nucleic acid sequencing, are widely used in
microbial taxonomy (Macdonell and Colwell, 1985). In addition, rep-PCR technique was
also found to be reliable, reproducible and rapid in generating highly discriminated
fingerprints for bacterial identification and classification (Rademaker and de Bruijn,

1997). In contrast to the traditional classifications that are based mainly on a few

phenotypic properties, the molecular identification and typing methods lead to more

general and natural classifications of bacterial Species (Towner and Cockayne, 1993).

1.2 The Genus Xanthomonas

Xanthomonas species are plant-associated bacteria. Xanthomonas infections occur
on at least 124 monocotyledonous and 268 dicotyledonous plant species (Leyns et al.,
1984). Some examples of the host plants are cotton (Zomorodian and Rudolph, 1993),
rice (Mew, 1993), mung bean (Vidaver, 1993), soybean (Hokawat and Rudolph, 1993),
citrus fruit (Stall and Civero, 1993), cabbage, broccoli, cauliflower (Schaad and alvarez,
1993), forage grasses (Leyns, 1993), tomato, pepper (Stall, 1993), apricot, cherry,
almond, peach, nectarine, plum (Civerolo and Hattingh, 1993), poplar (Ride, 1993),
strawberry (Rat, 1993), sugarcane (Rott, 1993), barley, rye, wheat (Duveiller and
Maraite, 1993) and mango (Pruvost and Manicom, 1993). Most of these are important
agricultural crops and fruits grown in different parts of the world. The disease symptoms
include necrosis, gummosis and vascular or parenchymatous diseases on leaves, stems or
fruits of the plants. The disease may cause significant loss in yield (Hayward, 1993).

Xanthomonas cells are Gram-negative rods, measuring 0.2-0.6].tm by 0.8-2.9um
(Swings et al., 1993). When cultivated on common growth media, most Xanthomonas
strains form yellow, water-insoluble pigments called Xanthomonadins (Swings et al.,
1993). Most Xanthomonas species have a polar flagellum and are strictly aerobic
chemoorganotrophs (Leyns et al., 1984). The genus Xanthomonas was proposed by
Dowson in 1939 and is classified under the superkingdom Prokaryota, kingdom Monera,
subkingdom Eubacteria, division Gracilicutes, phylum Proteobacteria and subclass

Gamma (Margulis and Schwartz, 1998; Stackebrandt et al., 1988). Seven species are

recognized within the genus Xanthomonas, namely: X. albilineans, X. axonopodis, X.
campestris, X. fragariae, X. populi, X. maltophilia (not a plant pathogenic species) and X.
oryzae. Nearly every new species and subspecies or pathogenic variant (pathovar) of
Xanthomonas was named after the host plant (Vauterin et al., 1993). As a result, some of
classifications may not reflect genomic relationships. Xanthomonas was reclassified by
Vauterin et a1. (1995) based on a comprehensive study of 183 strains using DNA-DNA
hybridization technique. The genus was shown to comprise 20 genomic species. Both the
genomic relationships and the needs of plant pathologists for a rational nomenclature are

taken into account in the new classification (Vauterin et al., 1995).

1.3 rep-PCR Genomic Fingerprinting

rep-PCR genomic fingerprinting is a DNA amplification-based technique for
generating bacterial fingerprints that can be used for species identification. It consists of
the following steps: (1) bacterial cells and DNA isolation, (2) repPCR amplification, (3)
gel electrophoresis, (4) gel image capture and (5) image processing and analysis
(Rademaker and de Bruijn, 1997).

Polymerase chain reaction (PCR) was devised by Mullis in 1984 for amplifying
specific DNA sequences (Watson et al., 1992). A PCR cycle consists of three stages,
namely, strand separation, hybridization of primers, and DNA synthesis (Stryer, 1995).
To prepare the mixture for PCR, two oligonucleotide primers, a DNA polymerase, and all
four deoxyribonucleoside triphosphates (dNTPs) are added to the target DNA. When
heated at a temperature of about 94°C for one minute, the double-stranded DNA
molecules separate completely, forming single strands that become the templates for the

primers and DNA polymerase. The solution is then cooled to about 50°C, a temperature

that allows the primers to anneal to the complementary sequences in the DNA molecules.
With the primed templates generated, the temperature is then raised to about 65°C, the
optimal temperature for the DNA polymerase. 65°C is maintained for eight minutes for
the DNA synthesis to proceed. These three steps are repeated for 30 cycles. All newly
synthesized strands serve as templates in each successive cycle; DNA sequences are
amplified as a result. The amplification is a billionfold after 30 cycles (Watson et al.,
1992; Stryer, 1995; Rademaker and de Bruijn, 1997).

In rep-PCR, DNA primers complementary to naturally occurring, highly
conserved repetitive DNA sequences present in the genomes of most Gram-negative and
several Gram-positive bacteria are used. They include three families of repetitive DNA
sequences, namely the 35-40 base pairs (bp) repetitive extragenic palindromic (REP)
sequence, the 124-127 bp enterobacterial repetitive intergenic consensus (ERIC)
sequence, and the 154 bp BOX element. The use of these primers in PCR leads to the
selective amplification of distinct genomic regions between REP, ERIC or BOX
elements. The collective name for this protocol is calledrep-PCR, and individually can
be referred to as REP-PCR, ERIC-PCR and BOX-PCR with respect to the different
primers (Rademaker and de Bruijn, 1997).

Since the amplified fragments of DNA are of different sizes, they can be resolved
in a gel matrix using gel electrophoresis to generate fingerprints (Rademaker and de
Bruijn, 1997). In gel electrophoresis, negatively charged DNA molecules move from the
negative pole toward the positive pole of the gel. The separation depends on the
molecular weight of the molecules. Smaller molecules move faster than larger molecules.

At the end of the gel electrophoresis, which usually runs for 18 hours, distinct bands of

DNA can be found (Hartl, 1994; Rademaker and de Bruijn, 1997). The rep-PCR genomic
fingerprints generated through PCR and gel electrophoresis permit differentiation to the
species, subspecies and strain level (Rademaker and de Bruijn, 1997).

The last Step of rep-PCR genomic fingerprinting involves making a hard copy of
the fingerprint. The gel is stained and a photograph of the gel is taken. The gel image is
processed and analyzed using numerical imaging tools (Rademaker and de Bruijn, 1997).
GelCompare (Applied Maths, Kortrijk, Belgium) is a software package designed for gel

image processing and analysis.

1.4 Digital Image Processing

Image digitization is the first step towards machine pattern recognition. Physical
images are converted into numerical representations by digitization. In the process, an
image is divided into small regions called pixels and at each pixel location, the brightness
of the image is sampled and quantized into a numerical gray level (Castleman, 1996). A
digital image can be seen as an array of different gray levels.

Digital images are used as inputs for pattern recognition systems. The design of a
pattern recognition system includes the following five steps, namely: image
segmentation, feature selection, classifier design, classifier training and performance
evaluation. Image segmentation involves the isolation of individual objects to be
recognized from the rest of the scene, using thresholding and edge detection techniques.
Feature selection involves choosing object properties that can best distinguish the various
classes of objects among themselves (Castleman, 1996).

Several techniques can be used to enhance image quality and reduce undesired

noise. Mean filter is one of the Simplest linear filters. It is implemented by a local

10

averaging operation where the value of each pixel is replaced by the average of all the

values in the local neighborhood of size N (Jain et al., 1995):

i+M
h(i)=,-'.- Xfac) (1.1)
t=i-M

where h is the filtered image, f is the original image, i is the pixel location and M = L121].

The size of the neighborhood N is directly proportional to the amount of filtering.
However, there is a trade-off between noise reduction and image detail. Sharp detail in
the image is often lost and steep changes will be blurred into gradual changes as a result
of linear filtering (Jain et al., 1995).

Wavelet analysis offers another technique for noise reduction. Wavelet
multiresolution analysis divides the frequencies of an image into octave bands, and the
image can be analyzed at different scales (Strang and Nguyen, 1996). The wavelet de-
noising procedure consists of the following three steps: (1) multilevel wavelet
decomposition of the image, (2) detail coefficients thresholding at each level, and (3)
wavelet reconstruction based on the original approximation coefficients and modified
detail coefficients (Donoho, 1995).

Discrete wavelet transform involves the decomposition of an image space into
orthogonal subspaces (Strang and Nguyen, 1996). Let V0 3 V1 3 V2 3 D V; be a
sequence of J nested subspaces, from fine to coarse scales. At each level j, the low
frequency approximation of an image is captured in the scaling subspace Vj, and the high
frequency detail of an image is captured in the complement wavelet subspace Wj. Each
subspace V,--1 is the direct sum of the coarser subspace VJ- and its orthogonal complement
Wj. The multiresolution decomposition of V0, the image space, can be written as

V0=W|$W2$...$W1®Vj (1.2)

11

(Liang, 1997). A wavelet decomposition tree is Shown in Figure 1.1, where S is the
image, and Aj and Dj represent the approximations and details, respectively (Misiti et al.,

1996).

 

 

 

 

m.
{“7

Da

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 1.1 Wavelet Multilevel Decomposition Tree

Mn} and {ll/jg} form the orthonormal bases for the subspaces Vj and W-,
respectively. A family of scaling basis functions can be obtained through dilations and
translations of a scaling function as shown in Equation 1.3:

¢,-,. = 2"”¢(2"' x-k) (1.3)
where ¢ is the scaling function, x is the pixel location, and j and k are integers
representing the scales and positions (Misiti et al., 1996). Similarly, a family of wavelet
basis functions can be obtained as Shown in Equation 1.4:

V0,. = 2"’2W(2"' x-k) (1.4)
where III is the mother wavelet and x, j and k are as previously defined. Some examples of
wavelet families are Haar, Daubechies, Biorthogonal, Coiflets, Symlets, Morlet, Mexican
Hat and Meyer (Misiti et al., 1996). The wavelet basis functions are dually localized in

both time and frequency domains (Barclay and Bonner, 1997).

12

The approximation wavelet transform coefficients a“ are given by the inner
product <s, ¢j.k> and the detail wavelet transform coefficients c“ are given by the inner
product <s, $139, where s is the image (Antonini et al., 1992). The discrete wavelet

transform can be implemented as a filtering algorithm using the following equations:

Cch = zgn—Zkaj—lm (1.5)

“13k = Zhu-Zkaj—lm (1.6)

which demonstrate that the coefficients c“ can be obtained by convolving the coefficients
a“), with g." followed by decimation, in which only the even indexed elements are kept.
Similarly, the coefficients a“ can be obtained by convolving the coefficients am, with h-"
followed by decimation (Liang, 1997). g-" and h-" are high-pass and low pass filters,
respectively.

After the decomposition at each level, thresholding can be applied to high
frequency detail coefficients to suppress high frequency noise. With a selected threshold
value, any detail coefficient whose value is below the threshold will be set to zero
(Donoho and Johnstone, 1995). The image can then be reconstructed with the noise-
reduced wavelet transform coefficients using inverse wavelet transform. The

reconstruction is shown as follows:
aj-I.n = 2 ng-nchc + Eth-nach (1.7)
k k

which demonstrates that aj. 1,, can be obtained by upsampling (zero-padding) c“ and a“,
followed by convolving c“ and a“ with g. and h” respectively, and taking the sum of the
convolutions (Strang and Nguyen, 1996). g" and h" are the reconstruction high-pass and

low-pass filters, respectively. The image can be reconstructed by passing the wavelet

l3

transformed coefficients through a series of the same reconstruction filters (Misiti et al.,
1996).

AS compared with other noise-filtering techniques, wavelet analysis does not
reduce the sharpness of edges, as long as the wavelet transform of the noise is smaller

than the significant features in the image (Weaver et al., 1991).

1.5 Backpropagation Neural Network

The neural network approach has been applied successfully for pattern
classification in several areas of biological research. Examples of such applications are
the protein classification artificial neural system (Wu et al., 1992), backpropagation and
counter propagation neural networks for phylogenetic classification of ribosomal RNA
sequences (Wu and Shivakumar, 1994), classification of mass spectra (Curry and
Rumelhart, 1990), identification of marine phytoplankton species using flow cytometric
data (Body et al., 1994), and identification of gram-negative rods using phenotypic
characters (Schindler etal., 1994).

Inspired by the structure of a human brain, a BPN is made up of interconnected
basic processing units called neurons (Rumeth et al., 1994). The model of a neuron is
shown in Figure 1.2.

Xo=1

X1

 

 

x,

 

 

 

XP

 

Figure 1.2 Model of a Neuron

14

For the k"I neuron,

 

r
V, =2wijj (1.8)
j=0
o, =(p(v,‘)= _v (1.9)
1+e ‘

where bk is the bias, Xj is the jth input, ij is the connection weight between the j‘h input
and the neuron, vk is the activation potential, (p(.) is the activation function, and 01‘ is the
output of the neuron (Haykin, 1994). The bias represents the internal resting value of a
neuron (Rumelhart et al., 1994). The activation function defines the output of a neuron
with respect to the activation potential. The nonlinear sigmoid function shown in
Equation 1.9 is the most commonly used activation function in BPN (Haykin, 1994).

A BPN is formed by arranging the neurons into a multilayer feedforward network.
The structure of a BPN is shown in Figure 1.3. The BPN is made up of an input layer,
one or several hidden layers and an output layer (Haykin, 1994). Neurons in input, hidden
and output layers are called input units, hidden units and output units, respectively

(Rumelhart et al., 1994)

 

 

0 t1
0 tk
O tK
Input Hidden Output Target
Layer Layer Layer Output

Figure 1.3 Structure of a BPN

15

The purpose of the BPN training is to find a set of connection weights that
enables the network to perform the desired nonlinear mapping from input space to the
output space (Rumelhart et al., 1994; Skapura, 1996). The supervised training of a BPN
consists of two passes, a forward pass and a backward pass. In the forward pass, the input
Signals from the input units propagate through the layers of the network and generate the
actual network outputs at each of the output units. The error signal at the kth output unit

for the nu‘ training pattern is defined by
e,‘ (n)=t,‘ (n)-o,‘ (n) (1.10)
where t is the target output and o is the network output (Haykin, 1994).

The energy of the network is a function of the error signals, and it can be

expressed on pattem-by-pattem basis as:

K
E(n) #243,202) (1.11)

k=l
where K is the number of output units (Haykin, 1994). In the backward pass, the output
layer weights and hidden layer weights are adjusted based on the error signals so as to
rrrinirnize the energy of the network. The weight update is defined by the delta rule:

6501)
5W).- (n)

 

Awfi(n)=-77 +aiji(n—l) (1.12)

where n is the learning rate, or is the momentum, and the use of the minus Sign before the

gradient term accounts for gradient descent in weight space (Haykin, 1994). For an

output unit with sigmoidal activation function, the weight update equation is

wk]. (n +1) = wk]. (n) + a[w,q. (n) - wk]. (n -1)]+ 776, (n)0j (n) (1.13)

16

where 6,01) = e,(n)0,(n)[1- o, (n)]. Similarly, for a hidden unit with sigmoidal
activation function, the weight update equation is

wji(n+1) = w..(n)+a[wfl(n)-wj,.(n—l)]+n5j(n)o,.(n) (1.14)

[I

where 51.01) = oj(n)[1- o). (n)]z 6, (n)w,q. (n) , and 8;, and 8,- are the local gradients of
k

neurons in the output and hidden layers, respectively (Haykin, 1994). Randomly selected
training patterns are presented to the BPN at each training iteration. The training is run
repetitively until the mean square error over the entire training set converges to an
acceptable minimal value.

An alternative to the pattern mode training described above is the batch mode
training. One complete presentation of the entire training set in the BPN training process
is called an epoch, and in the batch mode, weights updating is performed after every
epoch instead of after every pattern (Haykin, 1994). The use of batch mode training
allows the true gradient to be calculated (Demuth and Beale, 1994).

The stopping criterion is critical in the training. A network trained too specifically
to the training data may not generalize well to untrained data sets. Cross-validation is a
common scheme that avoids the problem of overtraining (Rumelhart et al., 1994). In
cross-validation, a set of untrained testing data is presented to the network, following
each training iteration, to evaluate the generalization performance. The training is
stopped when an increase in generalization error is found (Schittenkopf et al., 1997).

Learning rate and momentum are two parameters that control BPN learning. The
learning rate determines the amount of weight adjustment. A high learning rate may
accelerate the learning, but it may also cause instability. The momentum is added to

stabilize the training and to keep the weight adjustment moving in a steady direction

17

(Haykin, 1994; Freeman, 1994). Vogl et al. (1988) introduced the method of adaptive
learning rate and momentum to accelerate the BPN convergence. Starting with a small
learning rate and large momentum, the learning rate is increased continuously by a small
factor when the error is decreasing smoothly. However, when there is a rise in the error,
the learning rate is decreased by a large factor and the momentum is set to zero. A faster
convergence and more stable learning can be achieved with this method.

The knowledge of a BPN is stored in the final connection weights obtained
through the training. With the final connection weights, the BPN can be used for pattern
recognition and classification. In the BPN application stage, only the forward pass is

required to generate the corresponding outputs.

1.6 Performance Evaluation

Evaluation of classifier performance is needed to estimate the classification
accuracy and expected error rates (Castleman, 1996). Classification errors include
misclassification error, false rejection error and false acceptance error. In addition to
evaluating the performance of the classifier, the quality of the training data is also being
evaluated in the process (Skapura, 1996). A testing set is needed for the evaluation. The
partitioning of the available data into two equal halves, one for training and one for
testing is the most common method. To improve classification accuracy, classifiers

trained on different features can be combined to yield better results (Cho, 1997).

18

1 .7 Systems Analysis
Whitten and Bentley (1998) introduced systems development life cycle, a
systematic and orderly approach to solving system problems. A diagram of the systems

development life cycle is shown in Figure 1.4 (Whitten and Bentley, 1998).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

solution be solved
Related problem to be solved
Support ’ Analysis
I Implementation New solution to
error to be fixed same problem promem
Implemented analysis
solution V and
L-[ Implementation N——-+ Design d—J solution
Acceptable
solution

Figure 1.4 Systems Development Life Cycle

The five general problem solving steps are planning, analysis, design,
implementation and support. Planning involves identifying the scope and boundary of the
problem and formulating the development strategy and goals. Analysis involves studying
and analyzing the problems, causes and effects, as well as identifying and analyzing the
requirements that must be fulfilled by any successful solution. Design involves designing
the solution. Implementation involves implementing the solution. And lastly, support
involves analyzing the implemented solution, refining the design, and implementing
improvements to the solution (W hitten and Bentley, 1998).

Wetherbe and Vitalari ( 1994) developed a problem-solvin g framework, PIECES,

consisting of six categories:

19

P - the need to improve Performance

I - the need to improve Information and data

E — the need to improve Economics and control costs

C — the need to improve Control or security

E — the need to improve Efficiency of people and processes

S - the need to improve Service
The PIECES framework outlined some of the underlying needs and challenges in system
development.

Structured analysis was one of the first formal strategies developed for systems
analysis of information systems and computer applications (Whitten and Bentley, 1998).
Modern structured analysis is a process-centered technique, using a series of process
models to represent the essential processes of a system along with inputs, outputs and
storage (Yourdon, 1989). Another widely used approach for systems analysis is
Information Engineering (IE). It evolved from structured analysis. IE is a data-centered,
but process-sensitive technique (Whitten and Bentley, 1998). Since information is a
product of data, the study and definition of data requirements are emphasized before
those of process and interface requirements. Both data models and process models similar
to those in structured analysis are used in IE (Whitten and Bentley, 1988).

Data modeling is a technique for organizing and documenting the data of a
system. It defines the requirements for a database. An entity relationship diagram (ERD)
is used to depict data in terms of the entities and relationships described by the data.

Some concepts and definitions related to data modeling are summarized in Table 1.1:

20

Table 1.1 Data Modeling Concepts and Definitions (W hitten and Bentley, 1998)

 

Term Definition

 

Entity a class of objects, events or concepts about which we need to
capture and store data.

Attribute a descriptive property or characteristic of an entity.
Relationship an association that exists between one or more entities.

Cardinality a definition of the minimum and maximum number of
occurrences of one entity for a single occurrence of the related
entity. Cardinality must be defined in both directions for every
relationship, since all relationships are bidirectional.

Primary key an attribute that assumes a unique value for each entity
instance.

Foreign key a primary key duplicated in another entity to identify instances
of a relationship.

 

Processes transform data into useful information. A process model depicts the
structure and flow of data through a system and the work performed by the system. It is
made up of the following four components, namely: external agent, process, data store,
and data flow. External agents provide the net input to a system and receive net outputs
from a system. A complex system can be decomposed into smaller subsystems for
improved communication, analysis and design. Through process modeling, the
interactions of a system with its environment and the interactions among processes within
the system are described (Whitten and Bentley, 1998).

The systems analysis, data modeling and process modeling methods can be

applied to the design of a genomic fingerprint classification system.

21

Chapter II

MATERIALS AND METHODS

2.1 Data Collection

The rep-PCR genomic fingerprints data used in this research were obtained from
MSU DOE-Plant Research Laboratory. There were three sets of 752 fingerprints from
376 bacterial strains, and they were generated using BOX, ERIC and REP primers in the
rep-PCR reactions. The data represented 89 Xanthomonas pathovars. Each bacterial strain
in the BOX, REP and ERIC data sets had a duplicate set of fingerprints generated from
two different assays. A list of the data is shown in Table A.1 of Appendix A.

The fingerprint gel images were digitized using a flatbed scanner and pre-
processed in GelCompar (version 3.1 Applied Maths, Kortrijk, Belgium, no endorsement
implied), a window-based gel image processing software. Individual tracks of bacterial
fingerprints, each representing a bacterial strain, were segmented from the gel image. The
individual fingerprints were normalized with respect to two reference tracks, and
background subtraction was performed to enhance image quality (Rademaker et al.,
1997). The normalized rep-PCR genomic fingerprints were exported from GelCompar.
Each fingerprint was represented by 400 data points, with each point having a gray level
ranging from 0 to 255. A fingerprint can be viewed as a densitometric curve. A sample

BOX-PCR genomic fingerprint is shown in Figure 2.1.

22

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. f: I It I
g ,5 . I III [VIVA ll.
. “Willis W] u . m

Figure 2.1 BOX-PCR Genomic Fingerprint (LMG 12141)

Furthermore, the BOX, ERIC and REP PCR genomic fingerprints of the same
bacterial strain were linearly combined to form the BER fingerprints. The BER
fingerprint therefore, of one bacterial strain was composed of 1200 points. Then, the 1200
points were compressed to 400 points by replacing every three neighboring points with

their arithmetic mean. A sample BER fingerprint is shown in Figure 2.2.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

g: .Iiiunl i ii]
:‘A‘Nw W1.) ”U at. at. Ll...

Data point

Figure 2.2 BER Fingerprint (LMG 12141)

The rep-PCR genomic fingerprint data were in ASCII format and the structure of

the data files can be found in Figure A.1 of Appendix A.

23

2.2 Systems Analysis
A systematic approach is needed to design and develop a rep-PCR genomic
fingerprint classification system. The development of the rep-PCR genomic fingerprint
classification system described here follows the systems development life cycle shown in
Figure 1.4. The ultimate goal of the rep-PCR genomic fingerprint classification system is
to identify bacterial pathovars and assign them to the corresponding pathovar
classifications defined in the system. In the planning stage, PIECES problem-solving
framework (W etherbe, 1994; Whitten and Bentley, 1998) was used for problem
identification, and the needs are summarized below:
(1) The need to improve performance
I The genomic fingerprint classification system should be fast and accurate. The
current system using cluster analysis method classifies 50 bacterial Strains in
14 seconds on a Pentium II 266 PC with 96MB of RAM.
I The need to improve information and data
I The inputs including pathovar names, strains, fingerprints, indexes and others
should be captured accurately and not redundantly.
I The outputs of the system should be in a useful format, with accurate,
necessary and relevant information.
I Data should be well organized and accessible.
I Fingerprints graphical visualization capability should be available.
(2) The need to improve economics and control costs
I The costs for building the system and implementing the system should not be

too high. The prices of the GelCompare basic software, the cluster analysis

24

module and the identification module (developed by Applied Maths, Kortrijk,
Belgium) are $2600, $1800 and $1400, respectively.
(3) The need to improve control or security
I Redundantly stored data should be consistent in different files or databases.
I Processing errors should be controlled.
(4) The need to improve efficiency of people and processes
I Routine data processing procedures should be automated.
(5) The need to improve service
I The system should produce accurate, consistent and reliable classifications.
I The system operating procedures should be easy to learn and use.
I The system should be flexible to adapt to new situations and change.
I The system should coordinate well with external packages.
I The system should be compatible with other systems.

In the development process, systems planning was followed by systems analysis.

The problems identified in the planning stage were studied and analyzed, and the data

and process requirements were defined with the use of data modeling and process

modeling.

2.2.1 Data modeling

An entity relationship diagram (ERD) of the rep-PCR genomic fingerprint data is

shown in Figure 2.3. The data entities are pathovar name, strain, BOX fingerprint, ERIC

fingerprint, REP fingerprint, target pathovar, non-target pathovar, BOX classification,

ERIC classification and REP classification. The attributes of each entity are listed in the

respective boxes in the diagram. PK denotes primary key and FK denotes foreign key.

25

The line connecting two entities asserts the existence of a relationship. The definitions of
entities, attributes, primary key and foreign key can be found in Table 1.1.

As more than one strain could belong to the same pathovar, the pathovar name
entity has a one-to-many (zero or more) relationship with the strain entity. Since the
BOX, ERIC and REP fingerprints were generated from the same bacterial strains by
using different primers in the PCR reactions and each strain could have more than one
fingerprint, the strain entity has one-to-many relationships with the BOX, ERIC and REP
fingerprint entities. The target pathovar entity represents a collection of pathovar
classifications that the classifier is trained to recognize. The non-target pathovar entity
represents a collection of pathovar classifications that the classifier is not trained to
recognize. The pathovar name entity has one-to-zero or one-to-one relationship with both
target pathovar and non-target pathovar entities. The BOX, ERIC and REP classification
entities represent selected subsets of fingerprints for BPN training and classification. The
attributes include filtered fingerprint and classification results. Each of the BOX, ERIC
and REP fingerprint entities has one-to-zero or one-to-one relationship with the
respective BOX, ERIC or REP classification entity.

A separate ERD for BER fingerprint data is shown in Figure 2.4. It was separated
from the BOX, ERIC and REP fingerprint data so that it could be analyzed
independently. The relationships among the different entities are similar to those in
Figure 2.3. The data model in Figure 2.4 can also be adapted to analyze BOX, ERIC and

REP fingerprint data independently.

26

 

 

 

 

 

 

 

 

 

 

 

 

acreage. mouse a 5285 3583?. Sam a." 2:9“.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

\ u
or: 0:32 .
as: 2

693.52 .

5659".

 

 

«098.52 K

a

 

 

“F

 

 

 

 

‘ wow. \ J 95:23.9 m0 _
5:85.»me . Beacons“. . 238.24
3320 . Q": 585 . 95:96.2
5“. 3.65m. Soc. . PO . .
0.“: x09... . +OIIIT CF: —. .5 0.9;. II vl||._u
2n: 2 8&0. Eta—Lone.“—
ceeoeaaeo f emu L
( mum K
N .20.) N J N J N J
cozeommmmmw. 5:935“. . n
. OE 59.5. . mam .
fififl. roll? E .62.. KIT mu“ ”an”. BIL E z
b .
2n: 2 ago. . 252
«atonemen—
cozeoEeeeE 5.5m .2658
( oEm \ F 0.5 L W K F k
W .owj \ J
COSGOEmmfl—U . enigma—LE .
mOmwSU . Q": 59% .
C_l avg—OE“— . _O _ gay X009: . VT
0...: x85.
2% 2 $30. 5283...“.
9.038.535 xom
( xom \ F k

a

W a
§oo> $50 .

or: 252 .
2n: 2 693.

.265}.

( “once... K

 

 

27

 

 

25:23.2

>569; WOIT

95:26.2

to 92.. ._.O||_.

 

 

 

\ so

cofiéamao
3320 .
5”. news“. .
xouc_.
or: 2 320.
22.3.3320

( mum K

 

 

screams“. mum .0 5285 2,2222% beam ta 2:2“.

 

\ J

E3592". .
0.": seem .

Engine—u.

( mum K

 

 

E SE. . Elli.

 

\ J

or: one: .
9% sea .

wollir

 

 

r 5...:

 

D J

oi osmz .
252

f .2653. L

[I

 

 

‘P-

 

\ J

or: 252 .
2n; 2

899-52 .

335?.
/ «098.62 K

A
a

W J
.o.oe> 820 .

one 252 .
2% 2 $93.

 

 

5652.

r 890... K

 

 

 

28

2.2.2 Process Modeling

Data flow diagrams (DFD) were used to depict the processes and data flows in the
rep-PCR genomic fingerprint classification system. The notations of DFD are shown in
Table 2.1. The DFD of the whole rep-PCR genomic fingerprint classification system is
depicted in Figure 2.5. The main processes in the system were data input, image filtering,
data partitioning, BPN training, BPN classifier testing and performance evaluation.
Detailed DFDS of each processing components are shown in the respective sections
following. All BOX, ERIC, REP and BER data went through the same processes in the

system.

Table 2.1 Notations of Data Flow Diagram (Whitten and Bently,1998)

 

Symbol Description

 

 

External Agent

 

 

 

 

 

 

 

 

r———-\
Process
L___5
Data Store (Database or Files)
""'—’ Data Flow
. Exclusive Or Junction

CI

And Junction

 

 

 

29

 

BOX
Data "

 

""T—‘l
ERIC
Data

 

 

 

 

 

Data Image
Input Filtering

 

REP
Data

 

 

 

 

Data Target Training] r Data
Testing Set LPartitIomng

 

 

lI
I

 

Test
Date
I l BPN
BPN . .
. . Classifier
TraImn .

lPathovar
lCIassifications

Periorrnanc
Evaluatione

Results

 

 

 

Figure 2.5 DFD of rep-PCR Genomic Fingerprint Classification System

30

2.3 Data Processing
Data processing involved the creation of the rep-PCR genomic fingerprint

database and noise reduction in the fingerprint images.

2.3.1 Data Input

A database in Microsoft Access format was created for the rep-PCR genomic
fingerprints based on the data model in Figure 2.3, using Visual Basic 5.0 (Microsoft,
Redmond, WA, no endorsement implied). The structure of the database is specified in
Table B] of Appendix B. The BOX, ERIC, and REP fingerprint data were read into the
database, and data were stored in the Name, Strain, BF ingerprint, EFingerprint and
RFingerprint tables, respectively. Similarly, a database was created for BER data based
on the data model in Figure 2.4. A model of the data input process representing any one

of BOX, ERIC, REP or BER data is shown in Figure 2.6.

 

 

Store data
into
database

I I

Pathovar Strain Fingerprint
Name

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 2.6 DFD of Data Input

31

2.3.2 Image Filtering
Some noise was present in the fingerprint images. Although BPNs are quite

tolerant of noise, noise reduction might still help to improve classification accuracy. Two
filters, namely mean and wavelet, were used. Mean filter was implemented using
equation 1.1, at a neighborhood size of 5. Wavelet filtering was implemented with Matlab
Wavelet ToolBox (The Math Works Inc, Natick, MA, no endorsement implied), using the
Daubechies db8 wavelet with decompositions at levels 3 and 4. The threshold for wavelet
coefficient thresholding was selected automatically using the minimax criterion (Misiti et

al., 1996). The scaling function and mother wavelet of db8 are shown in Figure 2.7.

 

 

 

 

 

 

 

 

 

 

 

mumps Waveletfmotionpsl
l I ' * - 1 _ ' '
0.5 - - 0.6 »
o _ o
.05 . - 41.5 - J
1 .1 - I
o 5 to 15 o o 10 15

Figure 2.7 Scaling Function and Wavelet Function of db8

A DFD of image filtering is shown in Figure 2.8. Data for BPN training and
testing were selected from the fingerprint database. The selected data were stored in a
text file. The filtering programs would read the file and generate a new file with filtered
fingerprints. The filtered fingerprints were then stored in the field Filtered under the
respective B/E/RClassification tables of the database. These filtered fingerprints were

used for BPN training and classification.

32

 

Fingerprint

[ Select Data ]

I

Selected
Fingerprint

Mean W3V9j'9t None
Filtering Filtering

I I

Filtered
Fingerprint

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 2.8 DFD of Image Filtering

2.4 BPN Classifier Design
The BPN classifier design involved the following procedures: (1) Determination
of pathovar classifications to be included in the classifier (target classifications), (2) the

division of data into the training set and the testing set, and (3) BPN training.

2.4.1 Data Partitioning
The target pathovar classifications were determined from the data. To account for
possible variations in the fingerprints of the same pathovar, only pathovars with more

than four strains were selected as target pathovars. Those with fewer than four bacterial

33

strains were set aside as non-target pathovars. There were a total of 63 target pathovars

and 26 non-target pathovars. The target and non—target pathovar names were stored in the

Target and Non-target tables in the database, respectively. Each of the fingerprints that

corresponded to the target pathovars was assigned a target vector represented in binary

form. The 11th target classification vector was expressed in terms of one 0.8 and nine 0.23

with the 0.8 in the nth position. Next, the target data were further divided into two equal

halves: Set 1 and Set 2. Set 1 and Set 2 were used alternatively as training set and testing

set for cross-validation.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Filtered
Fingerprint
' Define D
Target
LClassificationb
I .
Target Non-target
Partition into Find
Training/Testing , Non-target
Sets L Set
Training Testing Non-target
Set Set Set

 

 

 

 

 

 

 

 

 

Figure 2.9 DFD of Data Partitioning

The DFD of data partitioning is shown in Figure 2.9. The 752 bacterial strains in

each of BOX, ERIC, REP and BER data types were partitioned as follows: 349 strains as

training set, 349 strains as testing set and 54 strains as non-target set. The training set,

34

testing set and non-target set were stored in three separate files for the BPN. The non-

target set could be used to test the true rejection capability of the BPN classifier.

2.4.2 BPN Training

BPN Classifiers for BOX, ERIC, REP and BER fingerprints were trained
independently using the respective data sets. The fingerprints were linearly normalized to
the scale of [0,2] from [0, 255] by dividing each data point by 127.5. The BPN performed
better with input data in this scale. The BPN was a multilayer feed-forward network
composed of an input layer, a hidden layer and an output layer. A diagram of the BPN

architecture is shown in Figure 2.10.

 

01
0k
063
Bacterial Input Hidden Output Pathovar
Fingerprint Layer Layer Layer Classifications

Figure 2.10 The Architecture of a BPN Classifier

For the BOX, ERIC and REP BPNS, the input layer contained 360 input units
which corresponded to the 31St to 390‘h data points of the 400-point rep-PCR genomic
fingerprint. This segment represented DNA fragment sizes between 200 to 12,000 base
pairs. The first thirty and last ten points were omitted, as they contained no relevant

information. For the BER BPN, 387 input units corresponding to the 11th to 397m data

35

points were used, since the BER fingerprints were compressed versions of linearly
combined BOX, ERIC and REP fingerprints. The output layer was made up of 63 output
units, and each unit corresponded to a pathovar classification. The hidden layer had 120.
hidden units. The number was determined by trial and error. The size of the hidden layer
is important as it determines the capacity of the BPN to learn the different rep-PCR
genomic fingerprint patterns. Although the learning capacity increases with the number
of hidden units, too many hidden units may actually cause “overfitting”. A network
learned too specific to the training data will not perform well when presented with
unfamiliar data sets. Moreover, a larger network will require more computations and
training time. Thus, it is desirable to keep the network simple, while maintaining the
capacity to learn the correct patterns. The nonlinear sigmoidal logistic function was used
as the activation function for all hidden and output units.

Matlab Neural Network Toolbox (The Math Works Inc, N atick, MA, no
endorsement implied) was used to simulate the BPNs. The supervised BPN training
followed the backpropagation algorithm using the gradient descent method (Demuth and
Beale, 1994). Batch mode training was applied, and the connection weights were updated
after every epoch to minimize the errors between actual network outputs and desired
outputs. Adaptive momentum and learning rate were used in the training. The system
would increase the learning rate by a factor of 1.05 when the new training sum square
error (SSE) was found lower than the training SSE in the previous epoch. If, however, the
new training SSE was found greater or equal to the previous training SSE, the learning
rate would be decreased using a factor of 0.7 and the momentum would be set to 0. Using

an adaptive learning rate and momentum, the BPN converged fast. The status of the

36

training was determined by analyzing the training and generalization SSES. The
generalization SSE was found by presenting the testing data to the BPN, after every 1000
epochs. The number of 1000 was chosen so that the decision interval was large enough to
avoid any local minimum. To avoid overtraining, the BPN training was terminated when
the new generalization SSE was found greater than the previous generalization SSE. The
final connection weights of each trained BPN were stored in a Matlab binary file. They

were used to set up the respective BPN classifiers.

2.5 BPN Classifier Testing
The trained BPN classifiers were validated with the training data and tested with
the testing data. The BPN classifiers were implemented with Matlab Neural Network

ToolBox. The DFD of BPN classifier testing is Shown in Figure 2.11.

 

Test Data

I

BPN
Classifier

I

BPN
Output

I

[ Thresholding]

Classification

 

 

 

 

 

 

 

 

 

 

 

Figure 2.1 1 DFD of BPN Classifier Testing

37

When presented with a test fingerprint as input, the BPN classifier assigned a
corresponding pathovar classification. Each output unit of the BPN yielded an output
score for the identification it represented, ranging from 0 to 1, with 0.8 and above for
perfect match and 0.2 and below for no match at all. A threshold was selected for making
classification decisions. Test bacterial strains were assigned pathovar classifications
represented by output units that had output scores above the classification threshold. The
classification threshold used had an effect on the classification accuracy. A high
threshold would lead to low misclassification error rate and high rejection rate, while a
low threshold would lead to high nrisclassification error rate and low rejection rate. The
BPN classifiers were also tested for true rejection using fingerprints of non-target
pathovars. A fingerprint would be rejected if none of the output units gave an output
score greater than the threshold. The pathovar classifications identified by the BPNS were
stored in the field Classifications under the respective B/E/RClassification tables of the

database.

2.6 Performance Evaluation

The BPN classifications were verified with the desired classifications, and the
performance of the BPN was evaluated based on top-2 recognition rate, misclassification
error rate, false rejection error rate and false acceptance error rate. A tree representation
of the BPN performance analysis is Shown in Figure 2.12. A top-2 recognition rate was
used for the evaluation since the top 2 identifications were important for the biological
studies, and some pathovars of the same Species had very similar fingerprints.

Furthermore, results from BOX, ERIC and REP BPN classifiers were combined

to improve accuracy. The final classifications were based on majority votes on the

38

classifications identified by BOX, ERIC and REP BPN classifiers. No pathovar
classification was assigned to a bacterial strain when there was a tie in the votes. Times

required for the major processes were also recorded.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Test Date
Target Non-target
Pathovars Pathovars
(Trained classes) (Untrained Classes)
Recognized Misclassified False- False- True-
rejected accepted rejected

 

 

 

 

 

 

 

 

 

 

 

 

Figure 2.12 A Tree Representation of the BPN Classification Performance Analysis

2.7 Graphical User Interface

A graphical user interface was developed using Microsoft Visual Basic 5.0 based
on the data models and process models. The system responded to user-triggered events at
the interface. The interface provided linkages to databases, files, process modules, and
the external Matlab computation engine. Data could be searched and retrieved with the
use of Structured Query Language (SQL). Information could be displayed on the
interface in data tables or charts. Fingerprints could also be compared visually.

The rep-PCR genomic classification system was divided into four main

components, namely Data Processing, BPN Training, BPN Classification, and

39

Comparison and Analysis. The Data Processing component consisted of data input and
image filtering modules. The BPN Training component consisted of data partitioning and
BPN training modules. The BPN Classification component consisted of BPN classifier
testing and performance evaluation modules. And lastly, in the Comparison and Analysis
component, a module was developed so that results of different BPN classifiers could be
compared. With the use of Multiple-Document Interface (MDI), multiple instances of

these components could operate simultaneously.

4O

Chapter III

RESULTS AND DISCUSSION

3.1 Image Filtering
Fingerprint noise reduction was accomplished using the mean filter and wavelet
filters methods. An example of a BOX fingerprint smoothed with mean filter is shown in

Figure 3.1. The filtered fingerprint is compared with the original fingerprint in the figure.

 

 

 

 

 

 

Graylevel

 

 

 

 

 

Data point — own”
— Filtered

Figure 3.1 BOX Fingerprint (LMG 12141) filtered with Mean FIIter

From Figure 3.1, we can see that the sharp edges in the peaks were rounded as a
result of mean filtering. Examples of mean filtered REP and ERIC fingerprints are shown

in Figure 3.2 and Figure 3.3, respectively.

41

255

 

170

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

E

> .

0

a .5 In 4 III

(9 M W

. I V
300 400
. — Original
Data Int
p0 -— Filtered
Figure 3.2 ERIC Fingerprint (LMG 12141) filtered with Mean filter
255

E 170

i I

as as .

(9

O ‘A‘ '7 "'
0 100 200 300 400
. —- ongInaI
Data Int

p0 — Filtered

FIgure 3.3 REP Fingerprint (LMG 12141) filtered with Mean filter

The results of wavelet filtering are shown in figures 3.4-3.9. Figure 3.4 depicts the
reconstructed original image and its approximations and details at each level before
detail-coefficient-thresholding. The original image can be seen as a linear combination of
a4, d4, d3, d2, and d1. Figure 3.5 depicts the detail wavelet transform coefficients of a
BOX fingerprint (LMG 12141) and the associated thresholds at each of the four levels.

Most of the noise filtered was from the details at levels one and two.

42

Signal and Approximationm

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

100 200 300 400 160 260 360 400

Figure 3.4 Reconstructed Image, Approximation and Details before Thresholding
(BOX Fingerprint - LMG 121 41 with db8 Wavelet at Level 4 Decomposition)

 

 

 

 

 

 

 

    

 

 

-20» l
100 200 300 400

 

FIgure 3.5 Detail Coefficients and The Associated Threshold for Noise Reduction
(BOX Fingerprint - LMG 121 41 with db8 Wavelet at Level 4 Decomposition)

43

Examples of BOX, ERIC, REP and BER fingerprints filtered with db8 wavelet at

level three decomposition are depicted in Figures 3.6, 3.7, 3.8 and 3.9, respectively.

 

 

Graylevel

 

. III

 

A

 

 

 

 

IA
VIW

 

 

 

100 200

Data point

 

300 400

— Original
— Filtered

Figure 3.6 BOX Fingerprint (LMG 12141) Filtered with db8 Wavelet at Level 3 Decomposition

255

 

170

 

 

 

Graylevel

o -I

 

 

 

r

100 200

Data point

 

300 400

— Original
-— Filtered

Figure 3.7 ERIC Fingerprint (LMG 12141) Filtered with db8 Wavelet at Level 3 Decomposition

255

 

 

 

 

 

 

 

'5
5 170
E 85 N A An I‘
0 o A i U
l I
O 100 200
Datapoint

 

 

300 400

- OIIgnaI
- Filtered

Figure 3.8 REP Fingerprint (LMG 12141) Filtered with db8 Wavelet at Level 3 Decomposition

255

 

 

 

 

 

 

 

 

 

 

 

 

E‘ I I
2 170 - I n l
s .I I
<9 85 II r
0 - _
o 100 200 300 400
Data point _ m9“
-- Filtered

Figure 3.9 BER Fingerprint (LMG 12141) Filtered with db8 Wavelet at Level 3 Decomposition

Examples of BOX, ERIC and REP fingerprints filtered with db8 wavelet at level
four decomposition are depicted in Figures 3.10, 3.11 and 3.12, respectively. Wavelet

filtering at level four was more rigorous since more detail coefficients were thresholded

as compared with level three.

255

 

 

 

 

 

 

 

 

 

 

 

 

3’ I
5 170 I -
E" 85 r MAI A m§l .
‘9 ,. WI WW I W
0 100 200 300 400
Datapoint :23:

FIgure 3.10 BOX fingerprint (LMG 12141) Filtered with db8 Wavelet at Level 4 Decomposition

 

       

 

 

 

 

255
s I
3,, III I I.. III .__._III _
‘” ,_ MIRIMIIMH'IUAII. ,1.
O 100 200 300 400
Data point _°°9'"”
— Filtered

FIgure 3.11 ERIC fingerprint (LMG 12141) filtered with db8 Wavelet at Level 4 Decomposition

45

255

S
2 1 70
>
m
b 85
(5

0

 

 

 

 

 

 

II II
it. . I I. II II
JAI'VIWLJMWWILJN KJIIJL I

Figure 3.12 REP Fingerprint (LMG 12141) filtered with db8 Wavelet at level 4 Decomposition

3.2 BPN Training

Data available for training and testing were divided equally into two halves, Set 1

and Set 2. A summary of the results using data Set 1 for training and data Set 2 for testing

is shown in Table 3.1. A summary of results using data Set 2 for training and data Set 1

for testing is shown in Table 3.2.

Table 3.1 BPN Training Results (Set 1)

 

 

 

 

 

 

 

Data Starting Number Training Generalization
Index Filtering Learning of SSE SSE
rate EMhs

BOX BPN
DATOOB-1 None 0.005 3000 1 .85 28.29
DAT01B-1 Mean Filter 0.005 4000 2.01 25.82
DAT028-1 Wavelet db8, level 3 0.005 4000 1.58 27.12
DAT03B-1 Wavelet db8, level 4 0.005 3000 2.53 28.43
ERIC BPN
DATO0E-1 None 0.005 3000 2.86 35.43
DAT01E-1 Mean Filter 0.005 3000 4.25 33.98
DAT02E-1 Wavelet db8, level 3 0.005 3000 3.00 34.86
DAT03E-1 Wavelet db8, level 4 0.005 4000 2.36 35.52
REP BPN
DATO0R-1 None 0.005 3000 2.29 35.22
DAT01R-1 Mean Filter 0.005 3000 3.37 32.23
DAT02R-1 Wavelet db8, level 3 0.005 3000 2.40 34.38
DAT03R-1 Wavelet db8, level 4 0.005 3000 2.65 34.89
BER BPN
BER00-1 None 0.004 7000 0.57 20.89
BER01-1 Wavelet db8, level 3 0.004 6000 1.03 25.41

 

 

46

Table 3.2 BPN Training Results (Set 2)

 

 

 

 

 

 

 

 

Data Starting Number Training Generalization
Index Filtering Learning 01 SSE SSE
Rate Epochs

BOX BPN
DATOOB-2 None 0.005 3000 1 .70 28.65
DAT01B-2 Mean Filter 0.005 5000 1.52 26.47
DATOZB-Z Wavelet db8, level 3 0.005 3000 2.05 27.91
DATOSB-Z Wavelet db8, level 4 0.005 3000 2.49 29.42
ERIC BPN
DATO0E-2 None 0.005 3000 2.87 36.64
DAT01E-2 Mean Filter 0.005 3000 4.1 1 34.67
DAT02E-2 Wavelet db8, level 3 0.005 3000 2.91 35.37
DAT03E-2 Wavelet db8, level 4 0.005 3000 3.13 36.61
REP BPN
DATO0R-2 None 0.005 3000 2.29 33.33
DAT01R-2 Mean Filter 0.005 4000 2.55 30.21
DAT02R-2 Wavelet db8, level 3 0.005 4000 1.78 31.15
DAT03R-2 Wavelet db8, level 4 0.005 4000 1.88 31.37
BER BPN
BER00-2 None 0.004 9000 0.36 19.89
BER01-2 Wavelet db8, level 3 0.004 6000 0.92 25.03

 

A smaller starting learning rate of 0.004 was used for the BER BPNS. The BER
BPNS training did not converge at a starting leaning rate of 0.005. Because the BER
BPNS had a larger input layer and the networks were more complex compared with BOX,
ERIC and REP BPNS, a lower learning rate was employed to avoid settling into local
minima and to ensure convergence.

The number of training epochs was determined by the generalization sum square
error (SSE). The BPN trainings were terminated when the generalization SSEs started to
increase. Among the BOX, ERIC and REP BPNS, the number of training epochs was
3000 to 4000. The BER BPNS had the largest number of training epochs (6000 to 9000)
and the lowest training and generalization SSEs among the BPNS, suggesting that there
were fewer inconsistencies in the BER fingerprints. This was expected since the BER

fingerprints contained all the features in BOX, ERIC and REP fingerprints and had a

47

higher information content. Any deficiency that existed in BOX, ERIC or REP features
was offset by the contribution of other good features. A plot of DATOOB-l BPN training
and generalization SSE curves is shown in Figure 3.13. The gap between training and
generalization errors started to increase after about 200 epochs. The training of a BPN

took about four hours on a UND( mainframe computer.

 

10000

 

__ Training SSE
....... Generalization SSE

 

 

 

 

 

 

Error

0
.0.
o..-
0..
.".“.-.“.”Q“a...“a...no-o“coco-0.00.0.0.0-0”.“ououououoo

 

 

 

 

 

1 E U I I E T I I I U I I E I I I I I l 1 I I I I I I I I I I

03006009001200150018002100240027003000
Epoch

Figure 3.13 sonic-1 BPN Training and Generalization Sum Square Enors

3.3 BPN Classifier Testing

All BPN classifiers were validated with the training data. One hundred percent
top-1 recognition rates were achieved when tested with data on which the BPN classifiers
were trained. These results showed that the classifiers were capable of classifying
bacterial fingerprints. Next, the BPN classifiers were tested with the testing data to
evaluate the performance of the BPN classifiers on untrained data sets and estimate the
true recognition rate and classification error rates. The classification threshold was found

by trying different values. It was found that at a classification threshold of 0.45, a high

48

recognition rate could be achieved with some small penalty in misclassification and

rejection errors.

An example of BPN output scores for a correctly identified bacterial strain is

shown in Figure 3.14. It had a high output score at output unit 49, which represented the

49th bacterial pathovar classification. An example of BPN output scores for an

unrecognized bacterial strain is shown in Figure 3.15. As shown in the chart, no output

scores were greater than the classification threshold of 0.45. An example of BPN output

scores for a bacterial strain with two identifications is shown in Figure 3.16. As shown in

the chart, there were two output units with output scores greater than the threshold of

0.45.

Score

Score

 

 

 

 

 

 
   

F; I 1 fl .
14 7101316192225283134374043464952555861
OutputU'ilt

Figure 3.14 BPN Output Scores of a Bacterial Strain with an Unique
Identification (DATOOB-1 - LMG 471)

 

 

 

 

 

    

14 7101316192225283134374043464952555861
OuiputU'ilt

Figure 3.15 BPN Output Scores of a Bacterial Strain with No
Identification (DATOOB-1 - LMG 12141)

49

Score

 

 

 

 

Output

    

14 7101316192225283134374043464952555861

Unit

 

Figure 3.16 BPN Output Scores of a Bacterial Strain with Two
Identifications (DATOOB-1 - LMG 8684)

3.4 Performance Evaluation

The results of the BPN classifier performance evaluation are shown in Tables 3.3.

Table 3.3 BPN Classifiers Performance Evaluation

 

 

 

 

 

 

 

 

 

 

 

Top-2 Mlsclassification False Rejection False Acceptance
Data Recognition Error Rate Error Rate Error Rate
Index Rate (%) (%) (%) (%)
(Over 349 strains) (Over 349 strains) (Over 349 strains) (Over 54 strains)
1 2 Mean 1 2 Mean 1 2 Mean 1 2 Mean

BOX BPN
DATOOB 94 93 94 2.0 2.3 2.2 4.0 4.3 4.2 57 65 61
DAT01 B 95 96 96 2.0 2.6 2.3 2.6 1.7 2.2 67 61 64
DATOZB 95 95 95 2.0 2.3 2.2 2.6 2.3 2.4 59 61 60
DAToaB 94 94 94 2.6 3.2 2.9 3.2 2.6 2.9 63 63 63
ERIC BPN
DATOOE 92 93 92 2.3 2.9 2.6 6.0 4.3 5.2 52 65 58
DAT01 E 93 93 93 3.2 2.9 3.0 4.0 4.3 4.2 52 65 58
DAT02E 91 93 92 2.9 3.2 3.0 5.7 3.7 4.7 61 65 63
DAT03E 93 93 93 2.0 3.2 2.6 5.2 3.4 4.3 57 72 65
REP BPN
DATOOR 92 91 92 3.2 2.6 2.9 4.6 6.0 5.3 67 65 66
DAT01 R 93 93 93 2.9 3.2 3.0 4.0 4.0 4.0 67 65 66
DAT02R 93 92 92 2.6 2.6 2.6 4.9 5.7 5.3 65 80 72
DATOSR 92 92 92 3.2 4.6 3.9 4.9 3.4 4.2 63 65 64
BER BPN
BEROO 99 98 98 1.4 0.86 1.1 0.0 0.86 0.43 50 61 56
BER01 98 98 98 0.86 0.57 0.72 1.2 1.7 1.4 46 61 54

 

 

 

 

50

Table 3.3 shows that mean filtering (DATOIB/E/R) increased the
misclassification error rate and decreased the false rejection error rate. Since the
decreases in false rejection error rates (1.4% on the average) were greater than the
increases in misclassification error rates (0.2% on the average), there were net increases
in the top—2 recognition rates (1.2% on the average) in all BOX, ERIC and REP BPNS
trained with mean filtered data. Also, wavelet filtering (DAT02B/E/R and DAT03B/E/R),
when optimized at a Specific level of decomposition, either decreased or maintained the
misclassification error rates and decreased the false rejection error rates. Wavelet
decomposition at level 3 using db8 worked best for BOX (DAT02B) and REP (DAT02R)
data, while wavelet decomposition at level 4 using db8 worked best for ERIC (DAT03E)
data. For the BPNS trained with the best wavelet-filtered data, the average decreases in
misclassification and false rejection error rates were 0.10% and 0.86%, and the average
increase in top-2 recognition rate was 0.96%.

Among the BPNS, the BER BPNs performed the best with the highest average
top-2 recognition rate of 98% and lowest average misclassification error rate of 0.93%.
The BOX BPNS performed better than the ERIC and REP BPNS with an average top-2
recognition rate of 95% and average nrisclassification error rate of 2.4%. The average
top—2 recognition rates of ERIC and REP BPNS were 93% and 92%, respectively. The
average misclassification error rates of ERIC and REP BPN S were 2.8% and 3.1%,
respectively. The results showed that the BOX features and data were better than those of
ERIC and REP. About 3% of the test strains were assigned two pathovar classifications

by the BPNS.

51

The false acceptance error rates for all BPNS were high (54-72%) when tested
with fingerprints of non-target pathovars. Ninety six percent of the misclassified and false
accepted Strains were assigned to some closely related pathovars of the same species.
This was due to the fact that bacteria of the same species, but of different pathovar, still
shared some common features in the fingerprints.

To enhance classification accuracy, the classification results from BOX, ERIC
and REP BPNS were combined to give final classifications. Different combinations of
BOX, ERIC and REP BPN classifiers were tried and the performances of the combined

classifiers are shown in Table 3.4.

Table 3.4 Performance of Combined BPN Classifiers

 

 

 

 

 

 

Top 1 Misclassification False False
Classifiers Recognition Error Rate Rejection Acceptance
Rate Error Rate Error Rate
(%) (%L (%) (%)
DATOOB-1
DATOOB-1 95 1.2 3.7 43
DATO0R-1
DAT01 B-1
DAT01E-1 96 1.2 2.6 44
DAT01 R-1
DATOZB-1
DATOZE-I 95 0.86 4.0 43
DAT02R-1
DAT038-1
DAT03E-1 95 0.57 4.6 43
DAT03R-1
DATOZB-l
DAT03E-1 95 0.57 4.3 41
DAT02R-1

 

 

 

Since the final classifications were assigned based on majority votes, the
recognition rates represented top-1 recognition rates. Recognition rates increased and
both of the rrrisclassification and false acceptance error rates decreased. The

rrrisclassification error rates of the combined classifiers were lower than those of the BER

52

BPN classifier. When the BOX, ERIC and REP BPNS with the smallest misclassification
errors were combined, a top-1 recognition rate of 95% was achieved together with a
lower misclassification error of 0.57%.

Bacterial strains wrongly identified or rejected by the BEROl-l BPN are Shown in
Table 3.5. Bacterial strains wrongly identified or rejected by the DAT02B-1, DAT03E-1
and DAT02R-1 BPNS are shown in Tables 3.6, 3.7 and 3.8, respectively. Bacterial strains
wrongly identified or rejected by the combined DATOZB-l, DAT03E-1 and DAT02R-1
BPN classifiers are shown in Table 3.9. Bacterial strains listed in Table 3.5 were found in
Table 3.6, 3.7 and 3.8 as well. Most wrongly-identified bacterial strains were assigned to
a closely related pathovar of the same species. Bacterial Strains listed in these tables need
further examinations. The errors were due to variations caused by biological and physical
factors in the fingerprinting process. The variations may also be inherent in characteristic

of the bacterial species.

Table 3.5 Bacterial Strains Wrongly Identified or Rejected by BEROl-l BPN

 

 

 

 

 

 

 

 

 

Strain Target Classification BER BPN Classification
LMG 539 X. axonopodis axonopodis X. axonopodis vasculorum "
LMG 558 X. axonopodis cajani - +
LMG 7473 X. axonopodis cajani X. axonopodis glycines
LMG 8689 X. axonopodis vitians -
LMG 937 X. axonopodis vitians X. axonopodis malvaceamm
LMG 7392 X. translucens cerealis -
LMG 883 X. translucens secalis -

 

 

 

 

" misclassification error
+ false rejection error

53

Table 3.6 Bacterial Strains Wrongly Identified or Rejected by DAT02B-1 BPN

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Strain Target Classification BOX BPN Classification
LMG 558 X. axonopodis cajani X. axonopodis malvacearum
LMG 7473 X. axonopodis cajani X. axonopodis glycines
LMG 9175 X. axonopodis citri E X. axonopodis alfalfae
LMG 8128 X. axonopodis glycines -

LMG 7427 X. axonopodis malvacearum X. axonopodis vitians
LMG 7443 X. axonopodis ricini -

LMG 937 X. axonopodis vitians -

LMG 7516 X. campestris campestn’s -

LMG 8134 X. campestn’s raphani -

LMG 7392 X. translucens cereaiis -

LMG 891 X. translucens cerealis X. transiucens hordei
LMG 8279 X. translucens hordei X. transiucens undulosa
LMG 7507 X. translucens secalis -

LMG 883 X. translucens secaiis -

LMG 875 X. translucens translucens X. transiucens hordei
LMG 876 X. translucens translucens -

 

54

 

Table 3.7 Bacterial Strains Wrongly Identified or Rejected by DAT03E-l BPN

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Strain Target Classification ERIC BPN Classification
LMG 488 X. albilineans -
LMG 8020 ' X. axonopodis alfalfae' -
LMG 539 X. axonopodis axonopodis -
LMG 558 X. axonopodis cajani -
LMG 7473 X. axonopodis cajani -
LMG 9172 X. axonopodis citri E -
LMG 8128 X. axonopodis glycines -
LMG 761 X. axonopodis malvacearum X. axonopodis glycines
LMG 780t2 X. axonopodis manihotis X. axonopodis phaseoli
LMG 7455 X. axonopodis phaseoli -
LMG 821 X. axonopodis phaseoli X. axonopodis manihotis
LMG 8689 X. axonopodis vitians -
LMG 937 X. axonopodis vitians X. axonopodis malvacearum
LMG 7516 X. campestris campestn's -
LMG 6518 X. oryzae oryzae X. oryzae oryzicola
LMG 595 X. translucens graminis -
LMG 726 X. translucens graminis -
LMG 8279t1 X. translucens hordei -
LMG 882 X. translucens hordei -
LMG 716 X. translucens phlei -
LMG 5260t2 X. transiucens translucens X. translucens undulosa
LMG 875 X. translucens translucens -
LMG 885 X. translucens undulosa X. translucens translucens
LMG 8276 X. vasicola holcico/a -
LMG 902 X. vasicoia vasculorum -

 

55

 

Table 3.8 Bacterial Strains Wrongly Identified or Rejected by DAT02R-1 BPN

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Strain Target Classification REP BPN Classification
LMG 5402 X. arbon'cola poinsettiicola -
LMG 539 X. axonopodis axonopodis X. axonopodis vasculorum
LMG 7194 X. axonopodis begoniae -
LMG 558 X. axonopodis cajani -
LMG 7387t1 X. axonopodis cajani -
LMG 7473 X. axonopodis cajani X. axonopodis glycines
LMG 9176 X. axonopodis citri A -
LMG 9181 X. axonopodis citri C X. axonopodis phaseoli(fusc)
LMG 9182 X. axonopodis citri D X. axonopodis citn‘ C
LMG 7429 X. axonopodis malvacearum X. axonopodis vitians
LMG 761 X. axonopodis malvacearum X. axonopodis vitians
LMG 937 X. axonopodis vitians X. axonopodis maivacearum
LMG 6518 X. oryzae oryzae -
LMG 471 X. sacchari -
LMG 8684 X. theicola -
LMG 8685 X. theicola -
LMG 880 X. translucens cerealis" -
LMG 595 X. translucens graminis -
LMG 882 X. translucens hordei -
LMG 728 X. translucens poae -
LMG 7507 X. translucens secalis -
LMG 883 X. translucens secalis X. oryzae oryzae
LMG 876 X. translucens translucens -
LMG 885 X. translucens undulosa -
LMG 892 X. translucens undulosa X. translucens translucens
LMG 736t2 X. vasicola holcico/a -

 

56

 

 

 

 

68232:: .2 .X

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

- 38329282 .2 .x 63232:: .2 .X 62285222622 .2 .X 68252:: .2 .X «an E
- - 22822222822 .x 68230:: .2 .X «822.2222: .2 .X mwm 022.2
- 2282222282 .2 .x - 2622222 .2 .X 86622222622 .2 .X who 02..
«82:22:: .2 .X M2283 .2 .x
- omNbo .o .X 2626222 .2 .X - 6.22663 .2 .X now 05...
- - M22382 .X - 6.2286» .2 .X name 05...
. - - 26922.2 .X 2622022 .2 .x «mm 0.2..
- - - 6.2522265 .2 .X 222.2826. .2 .X mmm 0.2..
262222 2282222622 .x
- M22698 .2 .x 6.2222628 .2 .x .626: .2 .x 6.226628 .2 .X 5m Os:
- - 3852622622 .2 .X - 2266:2382 .2 .x mum Os...
- - 228.5220 .0 .X omNbo .o .x 6659.2 .0 .X 9mm 0.2.2
- 6.222me228 .6 .X - - 622266222260 .0 .X @an Os...
EEmoomémE .m .x 2222226682229: .6 .X 223268922622 .6 .x . 8&2: .m .x 2.8 022.2
- 2.26.22: .6 .X 82262.26. .6 .x EEmoomsEE .m .x EEmoomémE .m .X 5s 05...
- 822.2926 .m .X - - 822.2923 .m .X mm 5 05:
2.22.293 .m .x «0:292 o. .m .X - 8.2.6.225 .m .X .Emfio d .X MK: 0.2.2
- - - 22232226692262: .6 .X 222N265 .m .X mmm Gs:
- ED202589. .m .x - 22239296 .6 .X 5.830229% 5 .X mmm 0.2.2
86> brofiE 232 2222822820 2222622535 2222822320 2202285320 Seem
22225222320 6222“. meo2.<o Tmmoktd Tmmeko 2692. .

 

 

 

 

 

 

 

52256 2% 2mm 2.3 Cam .528 2.8388 3 328.3 a 32.22282 222823 25% 222208 .3 asap

57

Times required for the major processes in the rep-PCR genomic fingerprint
classification system are shown in Table 3.10. Data processing, BPN training and testing
can be done in nine hours for 752 strains of bacteria. Fingerprint filtering and
classification of 752 bacterial strains can be done in about 15 minutes with a trained BPN
classifier. The BPN classification speed (0.2 seconds per strain on Pentium 133 MHz
with 32MB RAM) was faster than the GelCompare software using cluster analysis (0.3

seconds per strain on Pentium II 266 MHz with 96MB RAM).

Table 3.10 Speed Performance of rep-PCR Genomic Fingerprint Classification System

 

 

 

 

 

 

 

 

 

 

 

 

 

Number of Time
Process Bacterial Required
Strains (P-133 32MB RAM Win 95)

Data Input 752 35 seconds
Mean Filterian 752 4 minutes 21 seconds
Wavelet Filtering 752 11 minutes 7 seconds
Target Classifications Determination 752 1 minute 38 seconds
Trainigq and Testing Sets Partition 752 50 seconds
BPN Training 349 4 to 8 hours (UNIX)
Classification 349 1 minute
Verification 349 22 seconds

 

The size of the training data file (349 strains) was 620 kilobytes, and the size of
the BPN connection weights file was 408 kilobytes, amounting to a 34% saving in the
storage space. The saving will be more significant when the number of training data is
increased. The size of the BPN connection weights will remain the same as long as the
network structure is kept unchanged. This is an advantage as compared to the cluster
analysis method Since increased number of training data per classification will improve
recognition rate, but will not affect the storage space size and the time required for a

trained BPN to perform the classification.

58

3.5 User Scenario

A snapshot of the graphical user interface of rep-PCR Genomic Fingerprint
Classification System is shown in Figure 3.17. Names of databases created were shown
on the upper part of the left panel. Opened databases and their associated components
were shown on the lower part of the left panel. A tree view structure was used to
represent the organization. Processes could be activated by clicking on the nodes of the
tree. Activated components were shown on the right side of the main window. Several

components could be opened at the same time.

icon-rem 'Sett'ng

2......___

Neud Network and Iiu'iig Data Paranoia":

 

FIgure 3.17 A Snapshot of rep-PCR Genomic fingerprint Classification System

59

The data input screen is shown in Figure 3.18. Selected fingerprints data file, in

the format shown in Figure A.1 of Appendix A, could be read into the database.

 

 
 
   

' Data Processing - datm [BOX]
Data Processing

Rim-:3:

Select data file
Look in: IE BPN Papa E] :35

E NNBEEAPsm
El Nnbmpsm
E Nnercapsjrrt
B Nnrepaps.txt

 

 

    

 

 

 

 

 

 

 

 

 

 

File Danie: [Nnboxapslxt I | Qpen j
-_2 Filesoflvpe: ITexlI'litl] E]

E] Open as read-only

 

 

 

Figure 3.18 Data Input Screen

Figure 3.19 depicts a data selection screen. Data could be selected for filtering
and classification using SQL queries. Figure 3.20 depicts an image filtering settings

screen. Different filtering techniques could be implemented.

 

7" Data Processing - dalOI [BOX]

Data Piece-33mg

L339; Date i Select / Edit I Filtering | Output I

 

 

 

 

[NameBI- BIXenthomonas axonopodis iic'ni

 

 

 

 

 

 

 

 

 

 

FIgure 3.19 Data Selection Screen

, 7’ DaIEHPIDCESSIanifidaIIII IIIIIX] "W

Data: Pieces-giro

Reed Dete- ~3;BCI7EEI; Filter—B9 I MEI]

 

 

 

......- ...... —
NeidiboiSize.N: m

 

 

 

 

 

 

 

 

Figure 3.20 Image Filtering Settings Screen

61

The filtering results were shown in the screen depicted in Figure 3.21.

Fingerprints could also be visualized as densitometric curves.

 

9 Data Processing - datDI [BOX] _IZI

Read Data sggflsditl al.,-IQ] mph—l

HE mus—um

) EDIE-42:1 l_:lE- EELS. IIITE E m}: IIITIE. LMGn12141“
. D:\IANSUB\GELSINT\BXE- 039th5 LMG12141'

  

 

 

 

 

 

IllrlGIaph D:\.IAIIS=UB\13ELS INHBXE: 003 INT Is

 

{'1
f} it lei LMGIZHI" — Orfinl

 

 

I A II II. I.
. llllililllh iiit. ll.ll
-2_-A.iii.lli\l'VlliiUltil/U TIMI

 

 

 

 

 

 

 

 

 

 

Figure 3.21 Filtering Output Screen

Target pathovar classifications were defined in the target pathovar screen Shown
in Figure 3.22. Fingerprints belong to target pathovar classifications were divided into
training and testing sets as shown in Figure 3.23. Non-target pathovars are defined in the
non-target pathovar screen shown Figure 3.24. Figure 3.25 depicts the BPN settings

screen. BPN training parameters could be set, and BPN training could be initiated.

62

Iguanas”; i'EIaiI’ITIIIDQI ’ l 7

i

t

I flaunt: ’iet‘.'.'LII‘I+ T: Jl‘llii‘fl

 

 

Target I Taking/Testing Sets I Norrtarget I NauaI Network [ Fleporll

 

 

 

 

 

 

 

 

 

 

 

Figure 3.22 Target Pathovar Screen

Iii HFIEIII};ESIEI [3527777777 H

l

‘49:.ILII “Jenner? T':II:IIIII

 

 

Training
Set

 

 

 

 

 

 

 

Figure 3.23 Training / Testing Sets Screen

63

 

 

 

 

 

 

 

 

 

 

 

 

Figure 3.24 Non-target Pathovar Screen

IENN ham—Inf; ilulfll ”film

e‘leL;I.;I'I Netsuyril T'LII‘IIII;

I Tugetl TIairing/Test‘ngSetsl Nmtargeti Neural Network

 

 

Fie Two: -E -m.
Train

Hidden Unis:

0&th Units:

Starting Learning Hate :

Max Epoch: 10000

Fingerprint range: from to

 

 

 

 

 

 

 

 

 

FIgure 3.25 BPN Settings Screen

BPN classifier setting screen is shown in Figure 3.26. Classification data could be

selected. Fingerprints were classified using Matlab 5.1 as a computation engine.

W flilassrliatmn . magi [80X]

CILIszifIcnt‘Qn
Classification Sett'ngl Neural Network Owl Classificationm I
CiassificationData lTesting El ldat018.txs l I 0‘,”in l

Neural Network md Training Data Paranetersz

 

 

 

 

 

 

 

 

 

 

 

 

 

FIgure 3.26 Classification Setting Screen

The BPN outputs were shown on the BPN output screen depicted in Figure 3.27.
The output scores could be shown on bar chart. Figure 3.28 depicts the pathovar
classifications analysis screen. Incorrect classifications were identified, and recognition
rate and error rates were calculated. Figure 3.29 depicts the classification comparison

screen. Results form different classifiers could be loaded and compared.

65

- datm [BOX]

Neual Network Dupe!

it’im’" are if "—"' '

 

 

 

 

 

FIgure 3.27 BPN Output Screen

66

Classrlrcatron - dalUl [86x17 7

(.lJJSSIflLL’It‘IQI‘I

 

 

I Classiication Sett'ngl Neural Network Uutputi Classification Andysi: I

R ' ed 334/ 348-95707.
Mmied 258?. Fleiected: 1.72%

 

 

 

 

 

 

 

 

 

 

 

FIgure 3.28 BPN Classification Analysis Screen

“ Comparison 1. Analysrs

(,g‘II'II'JLIInon 5.: ‘ ' ’
”FQIEIN—WTIJFWW c: .fi Fusion DataSot m
ldatOlB I ldat01E J datOlR I

 
   

 

 

 

 

 

 

 

 

 

 

 

 

FIgure 3.29 Classification Comparison Screen

67

Chapter IV

SUMMARY AND CONCLUSIONS

Bacterial species and pathovars identification and classification are crucial to
pathological and ecological studies. Better disease control strategies can be sought with
deeper understanding on the disease-causing bacteria. The rep-PCR genomic
fingerprinting technique is capable of generating highly discriminated DNA fingerprints
for bacterial identification. An efficient classification system will better facilitate the
DNA fingerprint classification and pathovar identification process.

A BPN based rep-PCR genomic fingerprint classification system was developed.
Three sets of Xanthomonas genomic fingerprints generated from rep-PCRs, using BOX,
ERIC and REP primers, were used in the research. In addition, a fourth set of BER
fingerprints was formed by linearly combining the BOX, ERIC and REP fingerprints.
Mean and wavelet filtering techniques were used to reduce high-frequency noise on the
fingerprints.

Several BPN classifiers were trained to identify 63 Xanthomonas pathovars using
the BOX, ERIC, REP and BER original fingerprints and filtered fingerprints. The
performances of the different BPN classifiers were evaluated. Both mean and wavelet
filtering helped improve the recognition rates. Wavelet filtering was better at reducing

misclassification error rates, and mean filtering was better at reducing false rejection

68

error rates. The average top-2 recognition rates of BOX, ERIC, REP and BER BPN

classifiers were 95%, 93%, 92% and 98%, respectively. Furthermore, classification

results of selected BOX, ERIC and REP BPNs were combined to yield more accurate

results. By combining the results of three BPN classifiers with the lowest

misclassification error rates, a top-1 recognition rate of 95% was achieved together with a

misclassification error rate of 0.57% and a false rejection rate of 4.3%.

Based on the results, the following conclusions are stated:

1.

Bacterial rep-PCR genomic fingerprints could be classified effectively and
efficiently at the pathovar level using BPN classifiers. The BPN-based
genomic fingerprint classification system developed could be used to identify
63 Xanthomonas pathovars.

Classification accuracy improved with both mean filtering and wavelet
denoising. Wavelet filtering was better at reducing misclassification error
rates, and mean filtering was better at reducing false rejection error rates.
The BER BPNS performed the best, followed by BOX BPNs, REP BPNS and
ERIC BPNS with the average top-2 recognition rates of 98%, 95%, 93% and
92%, respectively. Combined BOX, ERIC and REP classifiers improved
classification accuracy. Ninety six percent of the wrongly identified bacterial

strains were assigned to a closely related pathovar of the same species.

69

Chapter V

RECOMMENDATIONS

Listed below are some of the recommendations for future research and

application:

1.

Since the trained BPN classifier is very data dependent, the training data
should cover all the possible variations in the fingerprint patterns. More
training patterns per classification should be included in the training set.

The target classifications should be well-defined based on genomic features.
Since the cost of rnisidentification is higher than false rejection, the classifier
design should be aimed at minimizing misclassification error rate while
maintaining a low false rejection rate.

Analyze problems behind the wrongly identified or rejected bacterial strains.
Try other wavelets for wavelet filtering.

The methodology can be applied to other bacterial genus and species.
Improve the user interface to include more functionality for data analysis.
Make the BPN classifier available on-line so that it can be used by a larger

group of researchers.

70

APPENDICES

71

APPENDIX A

rep-PCR GENOMIC FINGERPRINT DATA

Table A.l List of rep-PCR Genomic Fingerprint Data

 

 

Pathovar Strain BOX Index ERIC Index REP Index BER index
Trainingl'l’eeting Set 1

X W LMG 452 BXE-13JNT 5 ECE—225JNT 22 RPE~135JNT 5 XBER4C1.NT 5
X m LMG 457 BXE—237JNT 2 EOE-229.11" 2 RPE—233JNT 2 XBER502.1NT 2
X W LMG 455 BXE—195.INT 27 ECE~202JNT 27 RPE-ZOOJNT 27 XBER1301JNT 27
X W LMG 490 BXE-255JNT 19 ECE—252JNT 19 RPE—255.INT 20 XBER1302JNT 19
X Mm LMG 454 BXE-OO2.iNT 10 EOE—0051M 10 RPE—O47JNT 13 XBER361.1NT10
X arbarlaah aaryha LMG 555 BXE—039JNT 2 EOE-014.11" 9 RPE~O1 7.1NT 9 XBER1C2JNT 9
X. Micah aaryiha LMG 559 BXE-003.iNT 10 ECE-OOBJNT 1O RPS-009.15” 10 XBER101 .INT 10
X. Micah calyh LMG $50 BXE-012.iNT 13 ECE-O14.iNT 13 RPE—017.iNT 13 XBER162JNT 13
X Micah W LMG 747T BXE—flbJNT 25 ECE-225JNT 4 RPE—135JNT 25 XBER4C1JN‘I’ 25
X araadaala W LMG 750 BXE—ZxJNT 17 ECE-231.iNT 17 RPS-235.547 17 XBER501JNT3
X. ottoman/1m LMG 751 BXE-inNT 23 EOE-177.11" 21 RPS—152.11" 25 XBER701JNT 23
X eradicate Ind-ride LMG ”45 BXE-mJNT 29 ECE-231.iNT 29 RPE—235JNT 29 XBER501 .iNT 15
X. araadaah W LMG 5045 BXE—207.1NT 22 EOE-205.11" 22 RPE-205JNT 22 XBER15C1.NT 22
X arbadaah Mada LMG N47 BXE-O12JNT 14 EOE-014.15" 14 RPE-O17JNT 14 XBER1C2JNT 14
X my pahaattliaah LMG 5402 5XE-207JNT 14 EOE—205.11" 14 RPS-205.15" 14 XBER15C1.NT 14
X ubariaah mm LMG 5403 BXE-012JNT 15 EOE—014.5" 15 RPE-O17JNT 15 XBER1C2JNT 15
X. ottoman marl LMG 7435 BXE-199JNT 5 EOE-205.11“ 5 RPE-201JNT 5 XBER14C1.NT 5
X ubarlaah pmd LMG 552 BXE-O12JNT 17 EOE—014.1NT 17 RPS-01 7.1M 17 XBER1C2JNT 17
X ubariaala prunl LMG 553 BXE—207JNT 25 EOE—205.15" 25 RPE-205JNT 25 XBER1501.NT 25
X amadaah mm! LMG 553R BXE—272JNT 4 EOE—272.947 4 RPE—272JNT 4 XBER1502JNT 4
X arbodaah prud LMG 554 BXE-207.INT 27 ECE~205JNT 27 RPE-205JNT 27 XBER1SC1JNT 27
X. meal: plum LMG 555 BXE~250JNT 19 EOE-254.1!" 19 RPE-255JNT 19 XBER1502JNT 19
X Micah prunl LMG $79 BXE-152.INT 7 ECE-179JNT 5 RPE-1 74.1NT 9 XBER5C2JNT 22
X ubariaah pmd LMG eeeo BXE-O12JNT 15 EOE-014i!" 15 RPE-O17JNT 15 XBER1C2JNT 15
X axmapade Ma LMG 495 BXE-135JNT 4 ECE—225JNT 24 RPE-135.lNT 4 XBER401JN‘I' 4
X maapade M LMG 497 BXE—225JNT 23 EOE-255.1147 23 PIPE-227.110 23 XBER4C2JNT 23
X W um. LMG ”1511-1 BXE-237JNT 15 EOE-175.1197 3 RPE—151JNT 3 XBER501 .1147 5
X unmade um. LMG ”1511-2 BXE-139JNT 7 ECE-229JNT 15 RPE-233JNT 15 XBERSC2JNT 17
X mmpada em LMG 5079 BXE-139JNT 25 ECE-1 75.1NT 21 RPE-151JNT 21 XBER4CZJNT 27
X wnapada M‘ LMG $19 BXE-243.iNT 14 EOE—245.15" 14 RPE-249.lNT 14 XBER11C2JNT 14
X wnmads Ma' LMG 5&0 BXE-1%.INT 25 ECE-194JNT 25 RPE—204JNT 25 XBER1 101.15" 25
X cramped: elicit-0' LMG 5020R BXE-255JNT 25 ECE-350JNT 25 RPE-255JNT 25 XBER1302.NT 25
X axanapada M' LMG WR BXE-222JNT 24 ECE-225.INT 5 RPE-224JNT 24 XBER1BC1JNT 24
X. mnapade axanapada LMG 53511T BXE-039JNT 7 EOE—014.1NT 25 FIFE—01 7.1NT 25 XBER1C2JNT 25
X exam axanapada LMG 539 BXE-150JNT 7 ECE-177.iNT 5 RPE~152JNT 9 XBER562.|NT25
X amnpade wrapade LMG 5&1! BXE-241JNT 15 ECE-244JNT 15 RPE-247JNT 15 XBER902JNT 15
X wnapada axanapads LMG 540 BXE-159JNT 17 ECE—175JNT 15 RPE—142JNT 17 XBER501 .INT 17
X. wrapads bauhhba LMG 545 BXE-243JNT 15 ECE-245JNT 15 PIPE-249.11" 15 XBER11C2.1NT 15
X axanapads beam LMG 545R BXE-222JNT 3 ECE-223JNT 3 RPE-224JNT 3 XBER17C2JNT 15
X wnapade bagafle LMG 551 BXE-243JNT 4 EOE-245.1NT 4 RPE-249JNT 4 XBER11C2JNT 4
X wrapads bagarha LMG 552 BXE-139.iNT 3 ECE—175JNT 25 RPE~151JNT 25 XBER501 .lNT 3
X. mnapads beganbe LMG 7175 BXE-257JNT 3 ECE-3OOJNT 20 RPE—255.INT 3 XBER12C2JNT 3
X wrapads beganba LMG 7155 BXE—192JNT 27 ECE~194JNT 27 RPE—204JNT 27 XBER11C1JNT 27
X We begarha LMG 7194 BXE—237JNT 10 ECE—229.INT 10 RPE—233JNT 1O XBERSCZJNT 10
X wnapads bagartlu LMG 7194B BXE-193JNT 12 EOE-195JNT 12 RPE-205.iNT 12 XBER12C1JNT 12
X wuapads beguiee LMG 7195 BXE-225JNT 25 EOE-225.15” 25 RPE~227JNT 25 XBERSC1 .lNT 24
X woman's beganbe LMG 7195RA BXE-193JN'I’ 4 ECE—195.lN‘I' 4 RPE~205JNT 4 XBER12¢1JNT 4
X unmade baganlae LMG 7225 BXE-305JNT 2 ECE-350JNT 2 RPE-342JNT 2 XBER1ZC2JNT 2
X exampada beganlaa LMG 7303 BXE-139JNT 5 EOE-175.1NT 23 RPE-151JNT 23 XBERSC1 .INT 5

72

Xauanmadabaganlae
Xaxanapadsbagadae
Xaxanapadabaganbe
Xmiapadsaeganlae
Xmiapadaaalani
Xuanapadeaalanl
Xaxanapadeaahnl
erapadscmA
XaxanapadsdtdA
Xamvapadedtrm
anpadechm
XumapadacMA
XaxanapadadflA
XwnapadsdaM
Xanmpadedam
XaxanapadedtrlA
XmiapadeatiA
erapadedtric
Xammpadscmc
Xaxanapadaamc
Xuanapadaamc
Xmiapadealtric
XwnapadadflD
Xmiapadedtno
XamnapadeaME
XamnapadsamE
anpadeduE
XamnapadsdflE
XaxanapadaatlE
XaxanapadaamE
XaxanapadadflE
XaxanapadedarE

erapadsmllmaeanun
Xlxanapademlvaaearum
Xamrapadamelvaaeamm
'anapadsmehraaeamm
Xanmpademanlhatla
Xananapadamnlhada
Xamnapadamanlhatb
Xumapadamanhau
Xuanapadamm
anapademmthada
erapadsmanlhaas
Xaxanapadsmanhaea
Xaxanapadamuihafla
Xuanapadspatafl
Xananapadepatelll
Xmaapadspafl
Xumapadaphuaal
Xmaapadephuaal

LMGm
LMG7585
LMGT801
04678018
LM0558
LMG738711
LMG7473
LMG881
LMG881R
LMG8850
”46354
LMG8857
LMG8178
LMG8321
LMG8885
LMG8888
LMG8871
LMGESS
“46358
LMG8181
LMG8854
messes
LMG8182
LMG8182R
“468180001
”46818030
LMG8182
LMG8172
M8174
LMG8175
LMG8175R
M8325
LMG885
LMG7484
LMGW
wanes
LMG74W
LMG712
menace
LMG8128
LMG8128
M7427
LMG7428
worm
LMG781
warez
M784
M788
M788
M771
M773
M778
LMGMI
LMGW
LMG782
LMG784
M81112
“40811129
LMG81113
LMG7455
meson

BXE-243.INT 17
BXE-138JNT 18
BXE-243JNT 12
BXE—188JNT 18
BXE-237JNT 18
BXE-138JNT 8
BXE-237JNT 13
BXE-183JNT 28
BXE-258JNT 12
BXE-138JNT 13
BXE-258JNT 10
BXE-138JNT 15
BXE-238JNT 3
BXE-188JNT 13
BXE—305JNT 14
BXE-1MJNT 23
BXE-258JNT 27
BXE-158JNT 7
BXE-258JNT 2
BXE-158JNT 22
BXE-258JNT 12
BXE—IMJNT 25
BXE-238JNT 8
BXE-188JNT 5
BXE-238JNT 22
BXE-18OJNT 4
BXE-238JNT 7
BXE-183JNT 27
BXE-257JNT 27
BXE-207JNT 2
BXE-258JNT 23
BXE-188.INT 2
DXE-238JNT 18
BXE-207JNT 18
BXE—238JNT 12
BXE-158JNT 20
BXE-238JNT 17
BXE-158JNT 5
BXE-238JN'I’ 8
BXE-207JNT 18
BXE-280.INT 18
BXE-18OJNT 17
BXE-270.INT 4
BXE-210JNT 25
BXE-238JNT 25
BXE~181JNT 4
BXE-238JNT 27
BXE-IBIJNT 8
BXE-241JNT 8
BXE—210JNT 15
BXE-238.INT 18
BXE-210JNT 18
BXE-270.INT 12
BXE-207JNT 8
BXE-270JNT 2
BXE-210JNT 24
BXE-288.INT 4
BXE-210JNT 28
BXE-270JNT 7
BXE-WJNT 28
BXE-258.INT 24

73

ECE—248JNT 17
ECE-175JNT 12
ECE-248JNT 12
ECEe-202JNT 18
ECE-228JNT 18
ECE—228JNT 20
ECE—228JNT 13
ECE-360JNT 12
ECE~381JNT 13
ECE—175JNT 15
ECE—380JNT 18
ECE-I75JNT 13
ECE-23OJNT 3
ECE—202JNT 13
ECE—380JNT 23
ECE-202.INT 23
ECE-380JNT 28
ECE-178JNT 8
ECE—381.INT 3
ECE—178JNT 21
ECE-380JNT 18
EOE—202JNT 25
EOE-23OJNT 8
ECE-202JNT 5
EOE-mJNT 22
ECE~1 77.1NT 2
ECE-230JNT 7
ECE-301.iNT 7
ECE-281.INT 27
ECE-208JNT 2
ECE-381JNT 18
ECE-202JNT 2
ECE-23OJNT 18
ECE-208JNT 18
EOE-2mJNT 12
EOE-178JNT 18
EOE-23M” 17
ECE~178JNT 4
ECE~2MJNT 8
ECE—208JNT 18
ECE—284.INT 18
ECE-177JNT 15
ECE-273JNT 4
ECE-212.INT 25
ECE—231JNT 25
ECE-ITTJNT 28
ECE~258JNT 12
EOE-178JNT 4
ECE-244.INT 8
ECE-212JNT 15
ECE-231JNT 18
ECE-212JNT 18
EOE-273JNT 12
ECE—208JNT 8
ECE~273.INT 2
ECE-212JNT 24
ECE-288JNT 4
ECE~212.INT 28
ECE-273JNT 7
ECE~008JNT 28
ECE-283.lNT 24

RPE-248JNT 17
RPE—181JNT 12
HPE—248JNT 12
RPE—2OOJNT 18
RPE-Z33JNT 18
RPE-135JNT 8
RPE~233.INT 13
RPEWJNT 10
RPE—287JNT 13
PIPE-181 .INT 15
RPE—288JNT 12
HPE—181JNT 13
RPE—254JNT 4
RPE-2OOJNT 13
HPE-385JNT 7
RPE—2OOJNT 23
BPE-(KWJNT 18
PIPE-“2.lNT 7
RPE-287.INT 3
RPS-“2.lNT 22
RPE-ZGBJNT 13
RPE-200JNT 25
RPE-234JNT 8
RPE—ZOOJNT 5
RPE—234JNT 22
RPE-182JNT 8
RPE—234JNT 7
BPE-ZOSJNT 27
RPE—285JNT 28
RPE-208JNT 2
RPE-287JNT 24
RPE—ZO5JNT 28
RPE—234JNT 18
RPE—MJNT 18
RPE—234JNT 12
PIPE-1 42.INT 2O
RPE-234JNT 18
FIFE—1 42.INT 5
RPE—234JNT 8
RPE-208JNT 18
RPE~288JNT 18
RPE—182JNT 20
RPE-275.INT 4
RPE~217JNT 24
RPE~235JNT 25
RPS-1 73.INT 4
RPE—235JNT 27
RPE-I73JNT 8
HPE-247JNT 8
RPE—Z17JNT 14
RPE-235JNT 18
RPE-217JNT 15
RPS-$8.1M 10
RPE-208JNT 8
RPE—Z75JNT 2
RPE-217JNT 23
FIFE-3851M 8
RPE—217JNT 25
RPE-275JNT 7
RPE—WQJNT 28
RPE~287JNT 25

XBER11C2JNT 17
XBEFISCLINT 18
XBERI 102.|NT 12
XBERI3CUNT 18
XBERSC2JNT 18
XBEFMCIJNT 8
XBERSCZJNT 13
XBER12C1 .INT 28
XBER‘I4C2JNT 12
XBERSCI .INT 13
XBER13C2JNT 10
XBERSCiJNT 15
XBERGCZJNT 3
XBER13CIJNT 13
XBER13C2JNT 22
XBER13CI .INT 23
X8EHl3C2JNT 27
XBERSC2JNT28
XBER14C2JNT2
XBEH8ClJNT22
XBERI3C2JNT 12
X85813C1JNT 25
XBEHGC2JNT 8
XBER13CIJNT 5
XBEFI802JNT 22
XBEHGC2JNT 23
XBER8C2JNT 7
XBEFI1201JNT 27
XBER12C2JNT 27
XBEH14C2JNT 22
XBERISCI .INT 3
XBER13CI .INT 2
XBERBC2JNT 18
XBERISCLINT 18
XBER8C2JNT 12
XBEH801JNT 20
XBERBCZINT 17
XBERSC2JNT27
XBEMC2JNT 8
XBER15CI .INT 18
XBER1502JNT 18
XBER7CI .INT 17
XBERI8C3JNT 4
XBERI8CI .INT 25
XBEFIBCI .INT 10
XBER7C2JNT 18
XBEH8CIJNT 13
XBER7C2JNT 23
XBERQC2JNT 8
XBERI8C1JNT 15
XBERBCI .INT 2
XBER1801JNT 18
XBER18C3.INT 12
XBERMCZJNT 28
XBER1803.lNT2
XBER18CIJNT 24
X85818C2JNT 4
XBER1601JNT 28
XBEFIIGCZIJNT 7
XBERICi .INT 28
XBERISCIJNT 4

xammpodsplw
XWM
Xwnopodspluuol
anpodsmsool
XaxonapodlspInseat
xnxonopodlspllasode)
Xwnopodsplusoollflusc)
anapodlsplnsoolifluoc)
anopodlspMuOIIMsc)
Xwnopodspmmmx)
Xwnapodaplnaeotmnc)
anopodbplnsooflmm)
anapodlsplnuolmsc)
Xaxonapodspfnaoolmec)
xamwopodsridnl
Xwnopodsrlaw
Xwnopodbm
anapodsrldd
Xnmvapodsridnl
Xmmpodsw
Xwnopodsunwlnd
xumapodstanwhd
anapodsvuadaun
xmaapodlsvaswlonm
Xwnopodsvascwonm
xumopodsvasadanm
Xwnopodsvascuomn
Xamnopodsnm
xmnapodlsvasictm
Xmmodsmlcabm
xammpodtsmlcam
xmuapodlsmlam
anmapadsmlcam
xmnmodsmlcnm
xammodlsmmm
Xmmodsvoslcam
xmaopodsmkam
xwmpodsveslcam
anopodsvaslcuom
xmnapodlsvmns
xamnapodovlm
xbmml

Xbroml

Xbruml

Km

xm

xuoml
mebubuau
xmmo
xmtflscamb
XWW
XWW
xmmmm
xmmpoetriscanpostrls
xmmpasmacanmis
XWGIW
Xanrposlrbaanpastrh
xmm
XWQM
xmbmm
XWW

”46&1
”4682111R
LM682112R
”46823
”46842
”467458
”467511
mamas
mucosa
”4682511
”4683711
”4683711R
”4683712
”46841
”467441
”467442
”467443
”467444
”46881
”4632
”46”
”46855
”46894
moans
unease
mom
”46W1R
manna
”46W
”46888
menus
meson
mason
”46810
”46913
”4681412
”46822
”46828
”468%
unseen
”46837
mam
”468288
meazeon
”468272
LM68272R
”468272RR
”46547
”467385
”46587
”46588
”46571
LM6571R
”46573
”46573R
”46583
”467518
LM67518RA
”467882
meaom
”46m

BXE-181JNT 5
BXE-257.INT 28
BXE-MSJNT 8
BXE-238JNT 8
BXE-207JNT 5
BXE-280JNT 3
BXE-18OJNT 13
BXE—258JNT 13
BXE-189JNT 8
BXE-238JNT 28
BXE-180JNT 13
BXE—305JNT 15
BXE-180JNT 14
BXE-239.1NT 2
BXE—1 88.1NT 15
BXE—258.1NT 1O
BXE—1 81 .INT 29
BXE-«BJNT 20
BXE-181JNT 28
BXE-258JNT 8
BXE-218JNT 7
BXE-270.INT 17
BXE—222JNT 2
BXE-271 .INT 8
BXE-182JNT 13
BXE-24OJNT 8
axe-19mm 22
BXE-271JNT 24
BXE-1 82.1NT 20
BXE-241JNT 14
BXE-MJNT 29
BXE-271JNT 10
BXE—218.INT 23
BXE—O13JNT 2
BXE-27OJNT 23
BXE~308JNT 3
BXE-222.INT 15
BXE-O1 3.1NT 3
BXE-218.INT 22
BXE-3MJNT 5
BXE-182JNT 14
BXE-225JNT 8
BXE-138JNT 18
BXE-289JNT 2
BXE-138.INT 17
BXE-225JNT 28
BXE-21OJNT 12
BXE-WJNT 15
BXE-OO1JNT 1O
BXE-243JNT 9
BXE-OO1JNT 11
BXE-237JNT 15
BXE-183JNT 20
BXE-243JNT 18
BXE~£2JNT 8
BXE-237JNT 4
BXE-139JNT 17
BXE-340.INT 2
BXE-182JNT 24
BXE-243JNT 13
BXE-192JNT 5

74

565-177.le 27
EOE-m1 .INT 4
ECE-281JNT 28
EOE-231.1” 8
ECE~208.INT 5
EOE-284.1147 3
EOE-203.1” 13
ECE-381JNT 14
ECE—203JNT 8
EOE-231.1” 28
EOE-177M 11
EOE—380.1” 24
ECE-177JNT 12
ECE-231JNT 2
ECE-203.INT 15
EOE-381JNT 12
EOE-178.18" 25
EOE-$1 .INT fl
EOE-178M 24
ECE-381JNT 7
EOE—219.18" 7
EOE—273.1NT 17
EOE-223.18" 2
EOE—274.18" 8
EOE-1 79.1NT 11
ECE~232.INT 8
ECE-209.INT 5
EOE-274.1” 24
EOE-478.18” 17
EOE-244.1NT 14
EOE-008.1” 29
EOE—274.15" 10
ECE—218JNT 23
EOE-01 5.1NT 2
EOE—273.1” 23
ECE-378JNT 21
EOE-223JNT 14
ECE-O31JNT 12
EOE-21 8.1NT 22
ECE-378JNT 23
EOE-1 78.1NT 12
ECE-255JNT 8
ECE-228JNT 10
EOE-378.1” 14
ECE~228JNT 12
ECE~2554NT 28
ECE-212JNT 12
EOE-O1 5.1NT 1 5
ECE—OO7JNT 18
ECE-248JNT 8
ECE-(XWJNT 17
EOE-229.111” 15
ECE~221JNT 13
ECE—248JNT 19
ECE-223JNT 8
EOE-229.18” 4
EOE-475.11" 11
ECE-281JNT 8
805—1 84.1NT 24
ECE—248JNT 13
EOE—194.1NT 5

RPE-173JNT 5
RPS-285.15" 27
MWJNT 8
RPS-2354M 8
RPS-213.1” 5
RPS-288.18" 3
RPE-201.INT 13
RPS-287.1"? 14
RPE-201JNT 8
RPS—2344M 28
RPE-182JNT 15
RPE-342JNT 14
RPE-1MJNT 18
RPS—2354M 2
RPE-201JNT 15
H’E—ZBZINT 12
RPS-174.15" 3
RPE-307JNT 24
FPS-174.1“ 2
RPE-287.1NT 7
RPE—220JNT 7
RPS—2'75JNT 17
RPS-224.1"? 2
RPS-278.1” 8
RPS-1 74.1NT 15
IVE-23.11" 8
RPS-205.15" 22
RPE—278JNT 24
RPE—174JNT 22
RPE-247JNT 14
RPS-“.INT 28
RPE—278.INT 10
RPE-220JNT 23
RPS-018.15" 2
RPS-275.1” 22
RPE-MJNT 15
RPS-224.1” 15
RPS-01 8.1NT 3
RPS-220.1“ 22
RPS-308.11“ 17
RPS-1 74.1NT 18
RPE-227JNT 8
RPE-135JNT 18
RPS-288.18" 2
RPE-135JNT 17
RPE-227JNT 28
RPE—217JNT 12
RPE-018JNT 15
RPE-010.INT 18
RPE-248.INT 8
RPE-O10.INT 17
RPE-233JNT 15
RPE-342.lNT 7
RPS-249.18" 18
RPE—224JNT 8
RPE—233.INT 4
RPE—181JNT 11
RPE—285JNT 8
RPE-ZO4JNT 24
RPE-248.INT 13
RPE-204JNT 5

XBER702.INT 18
XBER1 202.1NT 28
XBER1201JNT 23
XBER702.INT 9
XBER14C2JNT 25
XBER1SC2.1NT 3
XBER14C1JNT 13
XBER1 402.1NT 13
XBER14C1.INT 8
XBER701JNT 11
XBER701JNT 13
XBER13C2JNT 23
XBER701JNT 14
XBER702JNT 2
XBER14C1.1NT 15
XBER14C2.INT 1O
XBER801 .INT 28
XBER1502JNT 4
xaemm .INT 28
XBER14C2.INT 8
XBER1803.INT 18
XBER17C1JNT 5
XBER17C2.1NT 14
XBER1 7C2.1NT 8
XBER901JNT 13
XBER802JNT 8
XBER12C1JNT 22
XBER18C1JNT 12
XBER901JNT 20
XBER8¢2.INT 14
XBER1C1JNT 29
XBER17C2.INT 10
XBER17C1JNT 23
XBER202JNT 2
XBER18C3.1NT 23
EERWC1 .INT 18
XBER17C2JNT 27
XBER202JNT 3
XBER17C1JNT 22
XBER18C1JNT 13
XBER901JNT 14
XBER4C2JNT 8
XBER4C1JNT 18
XBER1862JNT 2
XBER4C1JNT 17
XBER402JNT 28
XBER1801JNT 12
XBER202JNT 15
XBER201JNT 18
XBER1 102.1NT 8
XBEFQC1 .INT 17
XBERSC2JNT 15
XBER12C1JNT 2O
XBER1 1C2.INT 18
XBER17CZJNT 18
XBERSCZJNT 4
XBERSC1 .INT 17
XBER1202.1NT 8
XBER11C1JNT 24
XBER1 1C2.INT 13
XBER1 1C1.INT 5

XWW
xmw
xmm
XWW
xmmcanm
XWW
Xmabunmb
Xamposmscanpufls
XWW
xcnmpootrbm
XWW
XWW
XWW
XWW'
xmm'
XWW‘
xmpesmscanmh'
xmtnslncaruo
xanmm
xcnmposb'lsrapfwwl
XWW
XWW
XWW
Km
1m
XW
xm
xm
Km
xm

xcodul

xaodbd
warblbo
Xanadu-o
Xam
xmm
xm
lawman-o
xm
xm
ling-rho
xm
xmm
Xhonammhadaao
xhonorumhodaru
xnammm
xmm
Xhomrummdulo
Xhomrumpelaw
xmmpolargonl
Xmeolargonll
Xhortorzunpolupodl
Xhonorumpelarpod
Xhonorumpelaryonil
Xhorrorumpelaryam
Xhatorumpelaw
Xhmarumpelami
xmvorumpelamod
Xhonommpelatyod
xhomnunpehw
xlaomrunpolaryom

mam
LMG“
WSW
LMGKJS1
LMGN55
mam
LMGNQO
LMG81M
LM68112
LM68119
LMG8119R
LMG8121
LMG8123
LMGS75
LMG575RA
LMGT514
LMG7514R
M7421
M74”
”467505
“488134
LMGW
LMG“
LMGQ48
M5284
LMG5270
LMGWOR
LMG87O
LMG870
LMGB73
LMG”
mam
memo
LMG“
M74”
LMG7481
LMGfifl
mam
mom
LMGTOBH
LMGM
LMG710
LMG733T
LMG733TR
M734
M7413
M7414
LMG“
LMG7312
LMG7314
LMG7314R
LMG7314RFI
LMG7315
LMG7318
LMG7317
LMG7354
LMG7358
0467585
LMG7890
M78908
LMGWN

BXE-243.INT 2O
EKG-1%.!” 22
BXE-WJNT 4
BXE-192JNT 14
BXE-243JNT 28
BXE-192JNT 2
BXE-243JNT 1O
BXE-1flJNT 8
BXE-243JNT 8
BXE-192JNT 15
8XE-271JNT 18
BXE-192JNT 4
8XE-243JNT 22
BXE-13JNT 9
BXE—257JNT 9
8XE-139JNT 22
BXE-257JNT 22
axe-001 .INT 12
BXE-OO4JNT 19
BXE-WJNT 21
BXE-MOJNT 3
BXE-1MJNT 19
BXE-259JNT 5
BXE-1NJNT 8
BXE-O12JNT 22
8X8-13JNT 3
BXE-257JNT 15
BXE-OO3JNT 19
BXE-259JNT 19
BXE—MJNT 21
BXE—24OJNT 3
BXE-WJNT 23
BXE-238JNT 9
8XE-193JNT 18
BXE-238JNT 4
BXE~1 ”.INT 7
BXE-259JNT 4
BXE-(XXJJNT 25
BXE-O12JNT 2
BXE-241JNT 2O
BXE—O12JNT 3
BXE-1 38.1NT 22
BXE-225JNT 2
BXE-1 59.1NT 21
BXE-238JNT 18
BXE-180JNT 8
BXE—238JNT 5
BXE—207JNT 17
BXE-239JNT 24
BXE-(XBJNT 4
BXE—225JNT 5
BXE-180JNT 18
BXE—239JNT 15
BXE~1 80.1NT 22
BXE-239JNT 28
BXE-21OJNT 18
BXE-239JNT 28
BXE-207JNT 20
BXE-272JNT 5
BXE-210JNT 3
BXE-259JNT 15

75

EOE-248.1” 20
ECE-194JNT 22
EOE-380.18" 17
ECE~194JNT 14
EOE—248JNT 28
EOE-194.1” 2
EOE-248.1?" 1O
EOE-194.!“ 8
EOE-248.!” 8
EOE-194.1” 15
EOE-274JNT 18
ECE-194JNT 4
ECE-248JNT 22
EOE-228.1” 19
EOE-281.1” 9
EOE-175.18" 8
EOE-281.18" 21
EOE—MINT 18
ECE-O15JNT 19
EOE-(”7.1m 21
EOE-015.18" 22
EOE-203M 19
ECE~$1 .INT 8
ECE-ZMJNT 8
EOE—014.18“ 22
ECE~228JNT 25
ECE—281JNT 14
EOE-008.1147 19
ECE~283JNT 19
ECE—OOBJNT 21
ECE-232JNT 3
ECE-(DBJNT 23
EOE-23).!” 9
EOE-195.18" 18
EOE-230M 4
EOE—202.18" 7
EOE-$1.1m 5
ECE-WGJNT 25
ace—031.1117 3
ECE-244JNT 20
ECE—O55JNT 2
EOE-228.1NT 7
EOE-255.1” 2
ECE-178JNT 2O
ECE-230JNT 18
EOE-230.1” 27
EOE-2%.!“ 5
ECE—208JNT 17
EOE-231.1” 24
ECE-OOBJNT 4
ECE-255JNT 5
ECE-177JNT 14
ECE-231JNT 15
ECE-177JNT 20
EOE—231 .INT 28
ECE-Z12JNT 18
EOE-231JNT 28
ECE-208JNT 20
EOE-272JNT 5
ECE—212JNT 3
ECE-283JNT 15

RPE-249JNT 20
RPE-ZO4JNT 22
RPS-”.INT 18
RPS—204.1?“ 14
RPS-249JNT 28
RPE—204JNT 2
RPE-249JNT 10
BPE-ZOUNT 8
BPE-Z49JNT 8
BPS-204.1“ 15
RPE-278JNT 18
RPE—204JNT 4
RPS-249.1!" 22
RPS-135JNT 9
RPE-285JNT 9
RPS-181 .INT 8
RPS-285.1?“ 23
RPE-O10JNT 18
RPE-O18JNT 19
RPE—O10JNT 21
RPS-018JNT 22
RPS-201M" 19
RPS-342.11" 15
RPE-ZWJNT 8
RPS—017.1” 22
RPS-135JNT 3
RPS—285JNT 15
HPE-WJNT 19
RPE-287JNT 20
RPE-O17JNT 21
RPS-238.!” 3
HPE-MJNT 23
RPE-234JNT 9
RPE-205JNT 18
HPE—Z34JNT 4
RPS—200.1” 7
RPS—287.1” 5
RPE-MJNT 25
FIPE-O17JNT 2
RPE~247JNT 20
FIFE-017.1?“ 3
RPE—135JNT 22
RPE-227.INT 2
RPS-“2.lNT 21
HPE-234JNT 18
RPS-182JNT 11
RPE-234JNT 5
RPE-208JNT 17
RPE-235JNT 24
RPE-WJNT 4
RPS-227JNT 5
HPE-182JNT 18
RPE-235JNT 15
RPE—182JNT 24
RPE—235JNT 28
RPE—217JNT 17
HPE-ZSSJNT 28
RPS-208.1?" 20
RPS-272JNT 5
RPE-217JNT 3
RPE—287JNT 18

XBEFH 1Q.” 20
”581101.!” 22
XBER17C2JNT 12
XBER11C1.NT 14
”581102.11" 28
XBER11C1JNT 2
XBEH1 1CZJNT 10
X88811C1JNT8
X88311C2JNT 8
XBEH11C1JNT 15
XBER1BC1JNT 4
XBER1 1C1.1NT4
XBER11C2JNT 22
XBER4C1JNT 9
”831202.857 9
XBERSC1 .INT 22
XBER1202.INT22
XBER2C1JNT 18
XBEWJNT 19
”El-1201.1“ 21
XBER202JNT 22
XBER14C1JNT 19
XBER14C2.NT 5
”581301.017 8
XBER1C2JNT 22
XBER4C1JNT 3
XBER12C2.NT 15
XBER1C1JNT 19
XBER14C2JNT 19
XBER1C1 .INT 21
XBEMC2JNT 3
XBER1C1JNT 23
XBER802JNT 9
”591201.18" 18
mascara 4
”581301.” 7
memmm 4
XBER1C1JNT 25
XBER1C2.NT 2
XBER9-1JNT 4
XBEH1C2.NT 3
XBEFMC1 .INT 22
XBER4C2JNT2
XBER801 .INT 21
XBER802JNT 18
XBERBC2JNT 27
XBEMC2JNT 5
XBER1SC1JNT 17
”58801 .INT 9
XBER1C1JNT 4
XBER4C2JNT 5
XBER7C1JNT 18
X858702JNT 15
XBEFI7C1 .INT 22
XBER8C1JNT 14
XBER1801JNT 18
XBERBCLINT 12
XBER1SC1JNT 20
XBER18C2JNT 5
XBER1502JNT 23
XBER14C2JNT 15

xmw
xmm
xmunpolaryonl
xmw
Xhatarwnpolaw
Xhatanmpdaryonl
xmmw
Xhatommpolaryml
xmunpohmi
Xhom'umpehmonfi
xmmmm
Xhomrumvmum
Xhatorumm
xmm
xhammm
Xhomrunm
xmww
XW
thudlhl
lily-din!
KW
XWII
)0!”de
Xmabnb
um
xnuhnb
XJmIonb
xayuoayzu
Xmayuo
xayzuayzao
xaryuoatyno
Xmayao
Xmayuo
xayuoayuo
xayzaoayzlcoh
xaynooryzlcoh
Xmaydcah
Xmayzboh
Xayuooqzlcoh
Xmayzboh
Xaryzaooryzlcah
xayzuaryzlcoh
xaynoaydaah
X-PU

X4!“

in“

Xpapul

Xpopu

Xpapul

Xanadu!
Xanadu!
xm
xm
XJhdcoh
XJhdcoln
xnnummum
XWW
xmmamm
Kramhcansammhafi
XWW
xmmm

wanna
LMGT710

LMG 77108

LMG 7712

LMG 77128

LMG7715
LMGT783
LMG7784
LMG820

LMGB208
M7503
LMG7510
messes

mem1
mam

M938
LMG739'
M738
LMG740
LMG7419
LMG742
meson
LMG“
LMGWO
”46371
meson
LMGW4

LMG 50477
LMG 50471'8

LMG84111
0488518
LMG”
mama
LMG”
LM0857
LMG8578
LMGG81
LMG8818
LMG”
LM08858
LMG793
”467938
”40797

LMG 84711'
LMG 847118

W084712
LMG5743
LMG5753
LMG974
LMG471
LMG478
M4788
LMG“
LMG“
messes
LMGSOO
LMGSO1
LMG7278
LMG72711
LMG7384
LMG7392

BXE-207JNT 9
8XE—259JNT 18
BXE-207JNT 23
BXE-W5JNT 28
BXE-21OJNT 2
BXE-NSJNT 22
BXE-180JNT 18
BXE-239JNT 5
BXE-1 81 .INT 7
BXE~241JNT 17
BXE-181 .INT 22
8XE-240JNT 23
BXE-218JNT 12
BXE—271JNT 5
BXE-21 8.1NT 9
BXE—254JNT 2
BXE—138JNT 23
BXE-254JNT 3
BXE—180JNT 24
BXE—280JNT 28
BXE-MJNT 8
BXE-OWJNT 7
BXE-21OJNT 20
axe-305.041 24
BXE-181JNT 8
BXE-O13JNT 8
BXE-181JNT 18
BXE-013JNT 4
BXE-222JNT 18
BXE-270JNT 24
BXE—218JNT 18
BXE-240JNT 10
BXE-193JNT 8
BXE-241JNT 7
BXE-182JNT 12
BXE-257JNT 20
BXE-181JNT 24
BXE-257JNT 19
8XE-OO4JNT 5
BXE—240JNT 18
BXE-182JNT 5
BXE-305JNT 3
BXE-222JNT 9
BXE-O39JNT 12
BXE-218JNT 3
BXE-241JNT 2
BXE-(XJ3JNT 7
BXE-025JNT 2
BXE—003JNT 8
BXE-039JNT 29
BXE-139JNT 20
BXE-257JNT 18
BXE-OOZJNT 8
BXE-225JNT 13
BXE-138JNT 1 5
BXE-257.INT 4
BXE-138JNT 1 3
BXE—254JNT 5
8XE~004JNT 23
BXE-225JNT 18
BXE-138JNT 11

.76

EOE-208.1” 9
EOE-283JNT 18
EOE-208JNT 23
E08—381JNT 28
ECE-381JNT 27
505—264.INT 8
EOE-177.1“? 18
508-231.!” 5
EOE-178.847 3
EOE-244JNT 17
ECE-178JNT 17
EOE-232JNT 23
EOE-219.1” 12
EOE—274.18" 5
EOE-2191M 9
EOE—258.1” 14
EOE-228.” 8
EOE-258.847 15
EOE-177JNT 22
EOE-382.18" 3
ECE~008JNT 8
EOE-055.19“ 9
EOE-212.1NT 20
EOE-284JNT 10
EOE-178JNT 2
ECE—O13JNT 3
ECE-178JNT 12
EOE-058.1” 3
EOE-223.1NT 17
EOE-273JNT 24
EOE-219.18” 19
EOE—232JNT 10
ECE~195JNT 8
EOE-244JNT 7
ECE-179JNT 9
EOE—380.8" 7
605—1 78.1NT 20
ECE-380JNT 8
ECE—OO7JNT 5
ECE~232JNT 18
ECE-179JNT 3
ECE-380JNT 3
EOE~223JNT 9
EOE-015.18” 7
ECE~219JNT 3
ECE-244JNT 2
EOE-”.INT 7
ECE-O14JNT 8
ECE-WBJNT 8
EOE—055JNT 12
ECE-175JNT 8
EOE-$0.1m 5
ace-comm 8
ECE~255JNT 13
ECE-228JNT 14
EOE-”.INT 22
EOE-228.INT 15
ECE-258JNT 17
ECE-OO7JNT 23
ECE-255JNT 18
EOE-228JNT 17

RPS-”.INT 9
8P5-287JNT 19
RPS-“.INT 23
8PE-307JNT 28
8P8-217JNT 2
8PE-268JNT 8
8P5-182JNT 21
RPS-235.!” 5
8PE~173JNT 7
RPS-247.1” 17
8PE~173JNT 22
RPS-238.18" 24
8P5-220JNT 12
8P5—278JNT 5
RPE—220JNT 9
8PE-254JNT 8
8PE~135JNT 23
8PE-254JNT 7
8PE-182JNT 28
8PE-288JNT 28
8PE-047JNT 8
RPS-048.!” 7
8PE-217JNT 19
RPS-288.1“ 1O
8PE-173JNT 8
8P5—018JNT 8
8PE-173JNT 18
RPS-018.18" 4
8PE-224JNT 18
8P5—275JNT 23
RPS-220.18" 18
8P5-238JNT 10
8PE-205JNT 8
8PE-247JNT 7
8PE-174JNT 14
8P8-342JNT 8
8P5-173JNT 24
8PE—342JNT 5
8PE—010JNT 5
8PE-238JNT 17
8PE-1 74.1NT 7
8PE-342.INT 3
885-224%" 9
8PE-O18JNT 7
8PE~220JNT 3
8PE-247JNT 2
8PE-OOQJNT 7
8?E—017.lNT 8
8PE-OO9JNT 8
8PE-048JNT 9
8PE-181JNT 8
8PE-285JNT 18
8PE—048JNT 8
RPS-227.1“ 13
8PE-227JNT 14
8PE~265JNT 4
8PE-135JNT 13
8PE-254JNT 9
8PE-O1OJNT 23
8PE—227JNT 18
8PE~135JNT 11

XBE81402JNT 29
XBER1402JNT 18
X8681501JNT 23
XBE81502JNT 20
XBE81 502.1NT22
XBE81502.NT 8
XBER701JNT 18
”887021147 5
XBE8702JNT 22
XBE8902JNT 17
XBERBC1 .INT 2
XBE8901JNT 8
XBER1803JNT 25
XBE81702.NT 5
X8581803JNT 22
XBE8902JNT 18
XBE8401JNT 23
XBE8902JNT 19
X358701JNT 24
XBER1801JNT 8
XBER301JNT 8
XBE8302JNT 7
XBER1801JNT 20
XBE81502.INT 10
XBER7C2JNT20
X358202JNT 8
XBER801JNT 18
XBEWJNT 4
”581801.847 18
XBE81701JNT 11
”581701.847 18
”88802.18" 10
X3581201JNT 8
XBE8902JNT 7
XBER901JNT 12
XBE81202JNT 20
XBE8801JNT 24
X8881202.NT 19
XBER201JNT 5
XBE8901JNT 2
XBER802JNT 19
XBER1202JNT 8
X8E81702JNT 22
X858902JNT 2
XBEFR01 .INT 7
XBER17C1JNT 3
XBE8101JNT 7
XBER102JNT 8
XBE8101JNT 8
XBER302JNT 9
X358501 .INT 20
XBE81202JNT 18
XBE8301JNT 8
XBE8402JNT 13
XBE8401JNT 15
X8881202JNT 4
XBER4C1JNT 13
X888902JNT 22
X888201JNT 23
XBER402JNT 18
XBER401JNT 11

 

X um caulk LMG 887 8X5-237.INT 14 505—229JNT 14 RPS-233.11" 14 X858502JNT 14
X W M LMG 891 8X5-139JNT 4 EOE-175.1111“ 24 8P5-181JNT 24 X858501JNT 4
X. We candle' LMG 879 8X5—020JNT 5 505-015JNT 24 8P5-018JNT 24 X858202JNT 24
X um ”18' LMG 8798 8X5-241JNT 23 505-244.!NT 23 8P5-247JNT 23 X8589-1.INT 8

X W camllr LMG 880 8X5-225JNT 18 ECE-ZSSJNT 18 8?5-227.INT 18 X858402JNT 18
X W: gamble LMG 595 8X5-180JNT 5 505-177.!NT 3 8P5-182JNT 7 X858802JNT 24
X mm W LMG 713 8X5-280JNT 27 505-380JNT 18 8PE-306JNT 4 X8581801JNT 7
X W grunhls LMG 728 8X5-138JNT 28 505-228.!NT 3 8P5—1 35.1147 28 X858401JNT 28
X. “Mans gimbals LMG 7288 8X5-238JNT 2 505-25811“ 7 8P5~234JNT 2 X858802JNT 2
X um horde! LMG 737(8/7) 8X5-182JNT 18 505—179JNT 18 8P5-174JNT 21 X858901JNT 18
X. W hadal LMG 737(9I7) 8X5-241JNT 13 EOE-244.1117 13 885-247.11" 13 X858902JNT 13
X. “We harm! LMG W9 BXE-WJNT 27 EOE—COUNT 27 8P5-018JNT 27 X858201.INT 27
X mm W LMG 827911 8X5-271JNT 17 EOE-274.1141“ 17 885-278.11" 17 X8581801JNT 5
X. um horde! LMG 827912 8X5-210JNT 5 505-382JNT 2 885-217.1NT 5 X85815©2JNT 25
X. W8 horde! LMG 882 8X5-O20JNT 7 EOE-031.11" 24 8P5-O18JNT 28 X858202JNT 28
X W hadd LMG 884 8715-210.le 8 EOE—303.1147 2 885-217.11" 8 X8581502JNT 28
X tram pun! LMG 718 8X5-24OJNT 5 EOE—232.1117 5 8P5-238JNT 5 X858802.INT 5
X mm W LMG 719 8X5—181JNT 28 505-232.!” 12 8P5—173JNT 28 X858801.INT 28
X We pus! LMG 723 8X5-305JNT 17 EOE-378.1197 8 8P5-287.1NT 9 X8581402JNT 8
X W dial LMG 730 8X5—004JNT 28 505-015JNT 28 8P5-01 8.1NT 28 X858201JNT 28
X Wm LMG 594 8X5—280JNT 15 505-381.le 24 8P5-288JNT 15 X8581502.NT 15
X. um pan LMG 728 8X5-OO1JNT 4 ECE—OOBJNT 2 8P5-048JNT 2 X858301JNT 2

X am am LMG 7507 8X5-270JNT 18 505~273JNT 18 8P5-275JNT 18 X8581 701.1NT 8
X um soc-lb LMG sea 8X5-(X15JNT 3 EOE-018.817 3 8P5-048.INT 3 X858301.INT 3
X We mum LMG 5259 8X5-24OJNT 18 505—232.!NT 18 8P5-238JNT 19 X858901JNT 4
X “Mm mum LMG 520w 8X5—182JNT 8 505-179JNT 4 PIPE-174.118 8 X858802.INT 20
X W W LMG 528012 8X5-240JN1’ 8 505-232JNT 8 8P5-238JNT 8 )(858802JN‘I’ 8
X “mama mm» LMG 5282 BX5-222JNT 8 EOE—223.11" 8 8P5-224JNT 8 11858170288 20
X W m LMG 875 BX5-271JNT 23 EOE-382.11" 7 8135-385.le 10 X8581801.NT 11
X W m LMG 878 BXE—OMJNT 4 505-008JNT 4 8P5-048JNT 4 X858301JN'I’ 4
X tampons um LMG 8283 8X5-272JNT 3 505~272.INT 3 885—272.118 3 X8581802.lNT 3
X W Wuhan LMG 885 8X5—181.INT 18 EOE-178.11" 14 81’5—173JNT 18 X858801JNT 18
X um undubu LMG see 8X5—270JNT 18 505—273JNT 18 RPS-275.11" 18 X8581701JNT 4
X Ms W LMG 888 8X5-21 8.1NT 13 505-219JNT 13 8P5~220JNT 13 X8581803JNT 28
X mm m LMG m2 8X5-005JNT 5 505-018JNT 5 8P5—047JNT 7 X858302JNT 5
X vulcoh holdcoh LMG 73811 8X5—180JNT 9 505-177.INT 8 8P5~182JNT 12 X858802.INT28
X waded. holdaoh LMG 7311811 8X5-257JNT 14 505—300JNT 28 8P5-285JNT 14 X8581202.NT 14
X moon holdaola LMG 7381188 8X5-210JNT 9 505-212JNT 9 FIFE—217.11" 9 10581502.!" 29
X m. m LMG 73012 8X5-237JNT 25 EOE—229.11“ 24 8P5—233JNT 25 X858801JNT 3
X m holdcda LMG 73128 8X5—199JNT 2 505-208JNT 2 885-201JNT 2 X85814C1.NT 2
X vulcah holdcoll LMG 7418 8X5—013JNT 8 505-031JNT 14 8P5~018JNT8 X858202JNT 8
X W m LMG 7400 8XE-180JNT 2 505-178le 28 8P5~182JNT 4 1658802.!th 20
X ndcoh halclcoln LMG noon 8X5-34OJNT 3 505-378.INT 2 PIPE-307.1117 3 X8581202.NT 12
X unlock holdcoh LMG 8278 8X5-21OJNT 22 505-212.!NT 22 8PE-217JNT 21 X8581801JNT 22
X W holclaall LMG 82788 8X5-272JNT 8 505-272JNT 8 8P5—272JNT 8 X8581802JNT 8
X attach holcbob LMG 8277 8X5-210JNT 28 505—212JNT 28 8P5-217JNT 27 X8581801JNT 28
X waded: vasculamm LMG 8284 8X5-270JNT 28 505—273.INT 28 8P5-275JNT 28 X85817C1JNT 15
X W m LMG 900 8X5-182JNT 11 505~179JNT 8 8P5-174JNT 13 X858901JNT 11
X naval. newbmm LMG 902 8X5—241JNT 15 505-244.INT 15 8P5-247JNT 15 X858902JNT 15
X mm LMG 911 8X5-004JNT 9 505-055.INT 4 81’5—-010.INT 9 X858201JNT 9
X mm LMG 918 8X5-24OJNT 13 505-232JNT 13 885-238.11" 14 X858802JNT 13
X m LMG 917 8X5~182JNT 3 505-178.INT 27 8P5-174JNT 5 X858802JNT 17
X m LMG 919 8X5~271JNT 19 EOE-274.11“ 19 885~278.lNT 19 X8581801.|NT 7
X. mm LMG 92011 8X5—004JNT 10 505—055.INT 5 8P5-01OJNT 10 X858201JNT 10
X mm LMG 925 8X5-270JNT 27 505~273JNT 27 8P5~275JNT 27 X8581701JNT 14
X mm LMG 935 8X5~222.INT 17 505-223.!NT 18 8PE—224JNT 17 X8581702JNT 29
Training/Testing 8612

X. Minoan: LMG 482 BX5-225JNT 22 505—255.INT 22 8P5—227JNT 22 X858402JNT 22
X. 11mm LMG 487 8X5-139JNT 23 505-175JNT 5 885-181JNT 5 X858501JNT 23
X amen: LMG 488 8X5-258JNT 28 505—380.INT 29 8P5~288JNT 29 X8581 302.1NT 28
X. amen: LMG 490 8X5-198JNT 18 ECE~202.|NT 18 8P5—200JNT 18 X8581301JNT 18

77

XW
Xubomohcayflm
Xubaicolacaykn
)1.me
Xubaioohpglands
Xubodcolajuglands
Xubodoohjuglands
Xubodoolajwlm
Xubaicohjuglands
XWW
Xuborloolapolrm
XWW
Xabaicohplud
Xuborioolamn
Xuborioolapmnl
Xubodcohprud
Xarbodoobprw
Xubodcolaplml
Xuboricolnplud
Xubadoahpmnl
Xwnopodsm
Xwnopodsalhllao
Xuonapodam
thapodeam
anopodsallalho
Xmmpodsalfallu'
anapodsallallu‘
Xmmpodsdfalbe‘
Xwnapodsalhllu'
anopodsumopods
Xwnopodsaxonopods
Xwnopodlsmaopods
anopodlsaxmopods
Xmaopodabauhhlao
anapodlsbauhflu
Xammpodlsbogorm
Xumopodabogmbo
anapadsbegmho
Xwnapodsbogonlu
anpodsbogarm
anapodsbogaiu
Xwnopodabegavu
anopodlabeganho
anapodobogmho
Xwnapodlsbogmbo
mepodsbogadu
Xwnopodsbogam
anapodlsbegmo
Xaxonopodlsbagmho
anopodscunl
anonapodscnlanl
Xamnopodlscalanl
XamnopodlsdlflA
XwnopodsciMA
XwnopodiscmA
anompodldeA
anopodlsduA
XwnopodsdmA
anopoddeA
XwnapodscmA
XuonopodscmA

”46494
”46888
mecca
unseen
”46747T
”46750
”46751
menus
mam
”46W?
”46W
”4654M
”467438
”46852
”46853
”468538
”46854
”46855
menus
macaw
”46495
”46497
”468318114
mama-2
meson
”463119
mam
mama
mama
”4653811T
”46539
”465398
”46540
”46548
”465488
”46551
”46552
”467178
”467188
”467194
”4671948
”467198
”46719888
”467228
”46733
”467304
”467595
”467801
”4678018
”46558
”46738711
”467473
”46881
LM68818
macaw
”468854
”468857
”469178
”469321
”469885
”4698“

BXE-(XJ5JNT 1O
8X5—O25JNT 5
BX5-012JNT 1O
8X5-003JNT 13
8X5-225JNT 4
8X5-181JNT 3
8X5-239JNT 13
8X5-181JNT 15
axe-30mm 25
8X5—003JNT 14
8X5-280JNT 5
8X5-OO3JNT 15
8X5—259JNT 7
BXE-OOGJNT 17
8X5-280JNT 17
8X5-222JNT 21
8X5-280JNT 18
8X5-207JNT 28
8X5-240JNT 22
BXE-OOGJNT 18
8XE-225JNT 24
8X5-138JNT 5
8XE-139JNT 8
8X5-237.INT 17
8X5-225JNT 27
8X5—192.INT 13
8X5-243JNT 27
8X5-198JNT 24
8X5-272JNT 7
8X5-OO3JNT 28
8X5-238JNT 28
8X5-182.INT 24
8X5-238JNT 1O
8X5-192JNT 17
8X5-271JNT 15
8X5-192JNT 3
8X5-237JNT 22
8X5-193JNT 2
8X5-254.INT 9
8X5-139JNT 14
8X5-305JNT 4
8X5-139JNT 24
8X5-257JNT 5
8X5-1WJNT 29
8X5-237JNT 19
8X5-192JNT 18
8X5-237JNT 8
8X5-192JNT 1O
8X5-258.INT 17
8X5-139JNT 8
8X5-225JNT 20
8X5-139JNT 12
8X5-305JNT 7
8X5—199.INT 10
8X5-237JNT 12
8X5-198JNT 9
8X5-237.INT 9
8X5—159JNT 9
8X5-305JNT 12
8X5-198JNT 20
8X5~258JNT 24

78

505-018JNT 10
505—008JNT 9

505-01 4.INT 10
505-008.INT 13
505~255JNT 4

505-177JNT 25
505-231JNT 13
505-178JNT 11
505-381JNT 23
505-008JNT 14
505-264JNT 5

505-008JNT 15
505-381JNT 8

505~008JNT 17
505-381JNT 25
505—223JNT 20
505-284JNT 18
EOE-206.18" 28
505—232JNT 22
505-006JNT 18
505-255JNT 24
505-228JNT 23
505-175.INT 20
505-229JNT 17
505-228JNT 27
EOE-«1 94.INT 1 3
505-248JNT 27
505-202JNT 24
505—272JNT 7

505-008JNT 28
505-230JNT 28
505—179JNT 22
505~230JNT 10
505-1 94.INT 17
505-274JNT 15
505-194JNT 3

505u229.INT 22
505-19548” 2

505-248JNT 28
505-175JNT 14
505-380JNT 19
505-175JNT 4

505-300JNT 23
505-194JNT 29
505—229JNT 19
505-194.|NT 18
505—229JNT 8

505-1 94.INT 1 O
505-282JNT 17
505-175JNT 22
505-255JNT 20
505-175JNT 18
505-387.INT 2

505—203.!NT 10
505~229JNT 12
505~202JNT 9

505-229JNT 9

505~178JNT 8

505-380JNT 21
505—202JNT 20
505-380JNT 25

885-048.le 10
885-009.111” 9

8P5-017JNT 10
8P5—009JNT 13
8P5~227JNT 4

8PE-173JNT 3

8P5—235JNT 13
8P5-173JNT 15
8P5-307JNT 25
8P5-009JNT 14
885-288.1NT 5

8P5-OO9.INT 15
8P5-267JNT 8

8PE—009JNT 17
8P5-288JNT 17
8P5-224JNT 21
8PE—268JNT 18
8P5-208JNT 28
8P5—238JNT 23
8P5-009JNT 18
8P5-227JNT 24
8P5—135JNT 5

8P5-181JNT 20
885-233JNT 17
885-227.8" 27
8P5—204JNT 13
8P5—249JNT 27
8P5-2WJNT 24
8P5-272.INT 7

8P5—OO9.INT 28
8P5-234JNT 28
8P5~174JNT 28
885-234.1NT 10
8P5-204JNT 17
8P5-278.INT 15
8P5~204JNT 3

8P5—233JNT 22
8P5-205JNT 2

8P5-249JNT 28
8P5-181JNT 14
8P5-342JNT 4

8P5—181JNT 4

8P5-285JNT 5

8P5-204JNT 29
8P5-233JNT 19
8P5-204.INT 18
8P5—233JNT 8

8P5—204JNT 10
RPS—288.114"? 18
8P5-181 .INT 22
8P5-227JNT 2O
8PE~181JNT 18
8P5-342JNT 11
8PE-201JNT 10
8P5-233.INT 12
8P5-2MJNT 9

8P5-233JNT 9

8P5—234JNT 3

8P5~307JNT 14
8P5-200JNT 20
8P5—288JNT 25

X858302.INT 10
X858101.INT 9
X8581 02.1NT 1O
X858101.INT 13
X858402.INT 4
X858702.INT 17
X858702.INT 13
X858702JNT 29
X8581502JNT 13
X858101JNT 14
X8581 502.1NT 5
X858101JNT 15
X8581402.INT 7
X858101.INT 17
X8581502.INT 17
X8581801.INT 21
X8581502.INT 18
X8581501JNT 28
X858901.INT 7
X858101.INT 18
X858402JNT 24
X8584C1.INT 5
X8585C1.INT 8
X858502JNT 17
X858501JNT 25
X8581101.INT 13
X8581102JNT 27
X8581301.INT 24
X8581802.1NT 7
X858101.INT 28
X858701JNT 7
X858901JNT 24
X858802JNT 1O
X8581 101.1NT 17
X8581801JNT 3
X8581 101.INT 3
X858502.INT 22
X8581201JNT 2
X85811C2.INT 28
X858501.|NT 14
X8581202JNT 13
X858402JNT 28
X8581202.INT 5
X8581101.INT 29
X858502JNT 19
X8581 101.1NT 18
X858502JNT 8
X8581101JNT 10
X8581 302.1NT 17
X858501JNT 8
X858402JNT 20
X858501JNT 12
X8581202JNT 28
X8581401JNT 10
X858502.INT 12
X85813C1JNT 9
X858502JNT 9
X858801JNT 9
X8581 302.1NT 14
X8581301.INT 20
X8581 302.1NT 24

XaxmopodcdlflA
meopodsdlrlc
Xwnapodsdmc
Xmmpodlsdlrlc
eropodsduc
)Cmnopodsdmc
Kwnopodsdtrio
xmnopodsdwo
anopodscmE
KmnopodsclME
anopoddeE
XwnopodldeE
xmnopodlsdme
anopodscME
anapodlsdlrIE
XwnopodsduE
Kwnopodlsdaflonbadwu
anopodsdlefiamaflu
anopodsdeflbmdm
Xwnopodsdafimmr
Xmmpodsdomo'
Xwnopodsglycbm
anmopodsm
Kmnopodsm
Xwnopodsglychu
Xanmpodlsmahmcaarum
XaxonopodsmaMcumm
Xlxonopodsmahmmmm
X.axompodsma~acumm
anopodsmalncumm
anopodsmaMcumm
X.wnopodsmm
anopodsmam
xmnopodsmam
XWW
anopodsmmllm
Xwnopodsmam
anopodsmarmafls
Xmaapodsmanfim
anopodsmam
xmnopodspam
Xwnopodspam
anopodspam
anopodspfnaool
Xwnopodsmaool
anopodlsphauol
Xaxavapodsplnuol
anopodsphuool
Xwnapodlsplnuol
vaopodsphuaol
Xwnopodspmm)
anopodsphasecWhm)
Xwnopodsplnsoofimsc)
anopodsphaseollflwc)
X.uonopodspMsooli(hnc)
X.axonopodlspmsaoll(flm)
XenonopodlsphaseoWflm)
Xamnopodsmsodm)
anopodsphaaoomusc)
anopodsm
X.axonopodsrbld

LMG 9871
LMG $55
LMG N58
LMG 9181
LMG 9854
LMG 9858
LMG 9182
LMG 91828
LMG 9180001
LMG 9180¢ko
LMG 9182
LMG9172
LMG 9174
LMG 9175
LMG 91758
LMG 9325
LMG $5
LMG7484
mom
LMG 7399
LMG 7400
LMG712
mam
LMG 8128
LMG 8128
LMG 7427
LMG 7429
LMG 7430
LMG 781
LMG 782
LMG 784
LMG 788
LMG789
LMG771
LMG 773
LMG 778
memo"
LMG 78312
LMG 782
LMG 784
LMG 81112
LMG 811128
LMG 81113
LMG 7455
LMG N14
LMG 821
LMG 821118
LMG 821128
LMG 823
LMG 842
LMG 7458
LMG 7511
LMG M38
LMG m
LMG 82511
LMG 83711
LMG 837118
LMG 83712
LMG 841
LMG 7441
LMG 7442

8X5-1WJNT 28
8X5-237JNT 29
axe-19am 29
8X5—238JNT 15
8X5-198JNT 10
8X5-258JNT 28
8X5-159JNT 13
8X5-305JNT 9

8X5-180JNT 3

8X5~238JNT 23
8X5-1 59.1NT 14
8X5—257JNT 29
axe-19mm 25
8X5-259JNT 2
8X5-207JNT 3

8X5-258JNT 3

8X5-159JNT 25
8X5-305JNT 23
8X5-1 59.1NT 18
8X5-238JNT 13
8X5-159.INT 24
8X5—237JNT 27
8X5-159JNT 15
BX5-28OJNT 7

8X5-207JNT 25
8X5-239JNT 7

8X5-21OJNT 19
8X5-27OJNT 9

8X5-181JNT 10
8X5-239JNT 18
8X5-181.INT 13
8X5—239.INT 23
8X5-1 82.1741" 17
8X5—289.INT 8

8X5-181JNT 2

8X5-305JNT 28
8X5-210JNT 27
8X5-259JNT 28
8X5—210JNT 17
8X5-270JNT 8

8X5-21OJNT 13
8X5-270.INT 10
8X5-210JNT 23
8X5—012.INT 28
8X5-207JNT 4

8X5-239JNT 19
8X5-198JNT 24
BX5-340JNT 4

BX5-180JNT 20
8X5~259JNT 25
8X5-207JNT 12
8X5-259JNT 14
BXE-199JNT 12
8X5-259.INT 9

8X5-1 80.1NT 1 1
8X5-239JNT 3

8X5-1 98.1NT 2
BX5-239JNT 4

8X5-180JNT 12
8X5-259JNT 18
BXE-199JNT 9

79

505-22JNT 28
505-29JNT 29
505—202JNT 29
505-230JNT 15
505—202JNT 1O
505-380JNT 27
505-178JNT 12
505-282JNT 8
505-1 78.1NT 27
505—230JNT 23
505-178JNT 13
505—378JNT 4
505-209JNT 8
505-381JNT 18
505-208JNT 3
505—380JNT 14
505—178JNT 24
505-284.!NT 9
505—178JNT 17
EOE-22.1“? 13
505-1 78.1NT 23
505—229.!” 27
505-178JNT 14
505—284.!” 7
EOE-208.1147 25
505-231JNT 7
505-212.!NT 19
505-273JN'I' 9
505~178JNT 8
505-231JNT 18
505-231.INT 27
505-231.!NT 23
505-179JNT 15
505-289JNT 8
505-177JNT 24
505-387 .INT 8
505-212JNT 27
505-283JNT 28
505-212JNT 17
505-273JNT 8
505—212JNT 13
505-273JNT 10
505-212JNT 23
505-O31JNT 7
505~208JNT 4
505-231.!NT 19
505-209JNT 7
505-380JNT 11
505—177.le 17
505-283JNT 25
505-208.!NT 12
505-381.INT 15
505-203JNT 12
505~381JNT 11
505-177.INT 7
505-231JNT 3
505-202JNT 2
505-231JNT 4
505-177JNT 9
505-283JNT 18
505-203JNT 9

8P5—200.INT 28
8P5-2331NT 29
8PE—2MJNT 29
885-4341M 15
8P5-200JNT 10
885-288.18" 27
8P5-1 42.1NT 1 3
885-288M 7

8P5~182JNT 5

8P5—234JNT 23
885—142.1NT 1 4
8PE-307JNT 12
885-205JNT 25
885-287.1817 23
8P5—2OBJNT 3

885—288.1147 4

8P5—142JNT 25
885~288JNT 9

8P5-142JNT 18
885—234.1?" 13
885-442.1th 24
8P5-233JNT 27
885-142.1” 15
885—288.1?" 7

8P5—208JNT 25
885—235.1187 7

885-217JNT 18
8PE~275JNT 9

FIFE-173.1147 11
885-235JNT 18
885.473.le 13
885-235.18" 23
8?5-174.INT 20
885-308.1NT 8

8PE-173JNT 2

885-442.1817 18
8P5~21 7.1NT 28
885-287.1NT 27
8PE-21 7.1NT 18
885-275.1NT 8

8?5-308.INT 8

885-275JNT 10
8PE-217JNT 2
885-017JNT 28
885-208.INT 4

885-235JNT 19
885-205.INT 24
885-342JNT 9

885-182.1147 2
8P5-287JNT 28
8P5~208JNT 12
885~287JNT 15
885-201.1NT 12
885-281le 10
8P5~182.INT 13
885-235.1NT 3

885~200.IN'|' 2
8P5—235.INT 4

885-182JNT 14
8P5-287JNT 17
885-201JNT 9

X8581301JNT 28
X85M01JNT 7
X8581301JNT 29
X858802.INT 15
X8581301.INT 10
X8581302JNT 28
)(858801JNT 13
X8581302.INT 8
X858701.INT 4

X858802JNT 2
X858802.INT 7

X8581201.INT 27
X8581202JNT 27
X8581 402.1NT 2
X8581501JNT 3
X8581301JNT 2
X858801.WT 25
X8581502.lNT 9
X858801JNT 18
X85M02JNT 13
xasasm .INT 24
X858801JNT 5

X85M01JNT 15
X8581502JNT 7
X8581501JNT 25
X858702JNT 7
X85818C1JNT 19
X8581803.INT 9
X858702JNT 25
X858801JNT 4

X858702JNT 27
X858801JNT 8

X85201 .INT 17
X8581802JNT 8
X858702JNT 18
X8581802JNT 7
X8581801.” 27
X8581501.INT 8
X8581801JNT 17
X8581803JNT 8
X8581801JNT 13
X8581803.INT 10
X8581801.INT 23
X858102.|NT 28
X8581402.|NT 24
X858801.INT 5
X8581201JNT 24
X8581202JNT 25
$58701JNT 20
X8581 501.1NT 5
X8581 501.1NT 12
X8581402JNT 14
X8581401JNT 12
X8581402JNT 9
X858602JNT 29
X8587C2JNT 3

X8581301.INT 2
X858702.|NT 4

X858701JNT 12
X8581402.INT 18
X8581401.INT 9

xmm
xmm
xmnpodsm
Xaxonapodsdcld
mepodsum
anopodstamum
Xuonapodsmamn
Xaxmopodsvasculomm
)1.an
KWsmmm
Xnmnopodsvaswlorwn
Xwnopodsvnsculonun
anpodswslcnm
XWW
Xaxonopodsvaslcnm
)cwnapodlsvesiabda
Xmmodsmlcam
Xamwpodsmlam
Xmmpodsmlcam
Xmiopodsmlam
vaopodsmm
Xamnpodsmlam
Xwnopodsmm
Xmaopodsm
anpodsm
x3107»!

xbmml

Xbmml

X01031!

xbmml

11mm:
11mm
XWWO
XWW
XWW
XWW
XWM
XWW
XWWW
XWM
XWW
XWW
XWW
XWUBW
XWW
XWW
XWW
XWW
XWW
XWW
Xmmpostneanpafls

”407443
”407444
”4021
”40882
meson
”40955
”4024
”40895
”40”
”40901
”409018

”40827288
”40547
”407385
”40587
LM0588
”40571
”405718
”40573
”405738
”40583
”407518
”4075188A
LM07882
”403K”
mesons
mam
mamas
”408G358
LMGW51
”40m
”40m
1.1103099
”4081M
”408112
”408119
”1081198
”408121
”408123
”40575
”405758A
”407514

8X5-24OJNT 15
8X5-207JNT 13
8X5-24OJNT 14
BX5-199JNT 5
8X5-270.INT 19
8X5-218JNT 5
8X5-271JNT 14
8X5-218JNT 25
BX5-241JNT 5
8X5-181JNT 21
8X5-257JNT 24
8X5-22JNT 12
8X5-241JNT 12
8X5-182JNT 2
8X5-012JNT 29
8X5-21 8.1NT 28
8X5-271JNT 7
8X5-W4JNT 2
8X5-3MJNT 2
8X5-218.1NT 18
8X5-271JNT 27
BXE-CXMJNT 3
BX5-271JNT 8
8X5-22.|NT 13
8X5-241JNT 8
BX5-138.INT 19
8X5-25JNT 10
8X5-210JNT 10
8X5—25JNT 12
8X5-1%.1NT 2
8X5-289JNT 3
8X5-001JNT 9
8X5-004.INT 18
8X5-192JNT 8
8X5-004JNT 17
8X5-139JNT 1O
BXE-MSJNT 5
8X5-192.INT 18
BXE-271JNT 18
8X5-139JNT 21
8X5-237JNT 7
8X5-193.INT 5
8X5-243.|NT 25
8X5-12JNT 12
8X5-243JNT 8
8X5-192JNT 19
8X5-254JNT 7
8X5-218.INT 27
8X5-243JNT 15
8X5-192.|NT 25
8X5-243JNT 3
8X5-192.INT 9
8X5-243JNT 7
8X5-12JNT 7
8X5-243JNT 18
8X5-22JNT 4
8X5-243JNT 5
8X5-192JNT 20
8X5-25JNT 19
8X5-12JNT 8
8X5-237JNT 3

80

505-232JNT 15
505-208.|NT 13
505-232JNT 14
505-203JNT 5
505-273JNT 19
505—219JNT 5
505-274JNT 14
505-219JNT 25
505-244JNT 5
505-178JNT 18
505-380JNT 9
505-23JNT 11
505-244JNT 12
505-179JNT 20
505-014.INT 29
EOE-219.1147 28
505-274JNT 7
EOE-W7JN1'2
EOE-378.1147 19
505-219.|NT 18
505-274.INT 27
505-007.INT 3
505-274JNT 8
505-23JNT 12
505-244JNT 8
505-28.INT 9
505-255.IN'I' 10
505-212.INT 10
505-255JNT 12
505-28JNT 28
505-289JNT 3
ECE-OO7JNT 15
505-01 5.1NT 18
505-194JNT 8
505-01 5.1NT 17
ECE-175JNT 18
505-380JNT 8
505-194JNT 18
505-274JNT 18
505-175.INT 7
505-29.INT 7
505-1 95.1NT 5
EOE-248.1” 25
505-194JNT 12
505-248.INT 8
505-194.INT 19
505-248.INT 23
505-219JNT 27
505-248JNT 15
505-194JNT 25
505-248JNT 3
505-194JNT 9
505-248.INT 7
505-194JNT 7
505-248JNT 18
505-23JNT 4
505-248JNT 5
505-194JNT 20
505-255JNT 19
505-195.INT 8
505-229JNT 3

8P5-23JNT 18
885-”.INT 13
885-238.1NT 15
8P5-201JNT 5
885-275.!” 19
8P5-20JNT 5
885-278JNT 14
8PE-2OJNT 25
8P5-247JNT 5
8P5-173JNT 21
3173-0071117 7
8P5-24.|NT 12
885-247.1NT 12
8P5-174.INT 24
8P5-01 7.1NT 29
8P5-220JNT 28
885-278.111" 7
8P5—01OJNT 2
333-305.m7 12
885-220.1?" 18
8P5—278JNT 27
8P5-01OJNT 3
8P5-278JNT 8
885-24.!” 13
885-247.1” 8
8PE-135JNT 19
885-27.1917 10
8PE-217JNT 10
8P5-27JNT 12
8P5-135.1NT 2
8P5-289JNT 3
8P5-01OJNT 15
885-018.1NT 18
8PE-204JNT 8
8P5-018JNT 17
8P5-181JNT 18
8P5-342JNT 8
8P5-204JNT 18
885-278.1147 18
885—181.1NT 7
885-233.1” 7
8P5-205.INT 5
885-249.INT 25
885-204.047 12
8P5-249.|NT 8
885-204.!” 19
885-20JNT 27
885-249.1” 23
885-249.1NT 15
8PE-204JNT 25
8P5-249JNT 3
8P5-204JNT 9
885-249JNT 7
8P5-204JNT 7
8PE—249JNT 18
885-224JNT 4
8P5-249JNT 5
8P5-204.INT 20
885-227.1NT 19
8P5-205.INT 8
8P5-233JNT 3

x333302.m 15
3331501 .1117 1a
x333902.m714
11333140115175
x33317c1m77
1133315131117 17
XBER1BC1JNT2
113331701 111725
x3339c2.m75
x333301.1~7 21
11333120313724
x3331712m724
11333902137 12
x3339c1.1~722
x333102.13729
x33317c1.1~725
11333171211177
x333201.1~72
113331701 .INT 10
11333150311729
x3331ac1 .137 15
1133320111173
1133317021375
x33317121~725
113339021375
113334011117 19
x3334021~7 10
11333150113710
11333402111712
x3334c1.m72
11333190211173
x333201 .137 15
x333202.137 15
11333110111179
x333202.1~7 17
11333501 .137 10
11333121213723
113331 101.1117 19
11333130111175
1133350113721
1133350211177
x3331201.m75
1133311121117 25
1133311011117 12
1133311021375
x3331 101.137 19
x333111mu723
11333170113727
x3331 112.137 15
x3331101.m1'25
1133311021373
x33311c1.w79
x33311czw77
x33311c1.1~77
113331 102.117 15
x33317021~7 1e
x3331102m75
x3331101 13729
x3334c2.1~7 19
x3331201.m1'3
x3335021~73

xmhcanm'
xarrpourblncam
Xmmm
anpastrisraphui
xmmmu
xmmw
XWW
Km
Km
XM
)Cm
X00593“.
Xamm
xm
11w
11w
Xanadu.
11mm
11mm
K000113180
11mm
11%
Km
11m
11m
Km
11mm
Xhortorumlndaru
11mm
Xhonommhedomo
Xhonommhodarlo
xmm
xhonorumpolarymll
Xhorbmrnpalalyorfl
xmw
xhomrumpolalyodl
Xhomrwnpalamonll
Xhatorumpelelyonll
xmmnmpamgodl
11mm
Xhommmpelugoml
Xhatorumpolamml
11mm
xmm
xhwtormnpolamwl
XWMW
Xhodonmpolaryml
)1.me
XWUmpalaW
XWW
xmw
Xhonorwnpelargonl
Xhorvonunpelalyodl
Xhommmpelamodl
)LWnpquyoM
Xhonorumvim
11mm
xhortommvm
Xhommmmns
11mm
xmm

”4075148
”407421
”40742
”407505
”408134
”402011
”402012
”405248
”405284
”405270
”4052708
”40870
”40870
”40873
”40277
”40278
memo
mama
”40742
”407481
”40m1
”40282
meme
”40728
1.110705
”40710
”407331’
”4073318
”40734
”407413
”407414
”40285
”407312
”407314
”4073148
”40731488
”407318
”407318
”407317
”407354
”407358
”407585
”407890
”407%
”40772
”40772
”407710
”4077108
”407712
”4077128
”407715
”407783
”407784
”40820
”408208
”40752
”407510
”408888
”4029011
mam
”4028

8X5-193.INT 19
8X5-039JNT 13
axe-001.1117 13
8X5-020.INT 2

BXE-WJNT 2
8X5-25JNT 18
8XE-199JNT 4

8X5-258JNT 9

8X5-(D3JNT 2
8X5-25JNT 25
8X5-193JNT 14
BX5-012.INT 19
8X5-199.INT 18
8X5-012JNT 21
8X5-181JNT 17
8X5-012JNT 2
8X5-159JNT 18
8XE-257JNT 18
8X5-159.INT 11
8X5-258JNT 8

8X5-199JNT 3

8XE-012JNT 25
8X5-(XJ3JNT 2

8X5-254.1NT 4

8X5-(X)3.INT 3

8X5-25.|NT 7

8X5-138.1NT 27
8X5-238JNT 14
8X5-159JNT 23
8X5-238.INT 27
8X5-159.|NT 12
8X5-280JNT 8

8X5-181JNT 9

8X5-012.INT 4

8X5-138.INT 24
8X5-239.INT 8

8X5-180JNT 25
8X5-29JNT 12
8X5-181JNT 14
8X5-270.INT 3

8X5-181.INT 12
8X5—280JNT 12
8X5-22JNT 2
8X5-280JNT 23
8X5-199JNT 14
8X5—305.INT 19
8X5-199JNT 17
8X5-280JNT 14
8X5-207JNT 2
BX5-25JNT 27
8XE-207.INT 15
8X5-29JNT 8

8X5-180JNT 15
8X5-239JNT 2
8X5-182JNT 25
8X5-240JNT 7

8X5-182JNT 8

8X5-270JNT 25
8X5-218JNT 21
BX5-270JNT 2
8X5-241JNT 18

81

EOE-21.111“ 12
505-031JNT 2
505-007.INT 19
505-01 5.1NT 21
505-007JNT 2
505-381JNT 17
505—203.INT 4
505-282.INT 9
EOE-OMJNT 2
EOE-255.111" 25
505-195.INT 14
505-014JNT 19
505-203JNT 18
505-01 4.1NT 21
505-178.INT 13
505-01 4.1NT 2
505-178JNT 15
505-281JNT 17
505-178.1NT 9
EOE-20.1517 17
505-203.INT 3
505-014.INT 25
505-014JNT 2
505-258JNT 18
505-031.INT 4
EOE-255.1147 7
505-28JNT 2
505-22JNT 14
505-1 78.1N'l' 2
505-258JNT 10
505-178JNT 11
505-284.1NT 8
505-178JNT 5
505-014.INT 4
505-28JNT 5
505—21JNT 8
505-1T7.INT 23
505-21JNT 12
505-178.INT 9
505-273.INT 3
505-1 78.1NT 7
505-284.INT 12
505-23.INT 21
505-284JNT 23
505-203JNT 14
ECE-283.|NT 2
505-203.INT 17
505-284JNT 14
505-208JNT 2
505-378JNT 9
505-208JNT 15
505-231.INT 8
505-1 77.INT 13
505-231JNT 2
505-179.INT 2
505-232JNT 7
505-179JNT 8
505-273.INT 25
505-219JNT 21
505-273.INT 2
505-244JNT 18

RPS-22.1“ 19
8PE-018JNT 18
8PE—010JNT 19
8PE-018.INT 21
8PE-010JNT 2
8P5-27JNT 2
8PE-201JNT 4

8PE-288JNT 10
RPS-GDJNT 2
885-27.1!" 25
895-205.!” 14
8PE—01 7.1NT 19
885-201.!” 18
8PE-047JNT 15
"’5-173JNT 17
885-017.1NT 2
8P5-142JNT 18
8PE-285JNT 19
8PE-142JNT 11
885-27.111" 13
885-201.1117 3

885-017JNT 25
8P5—(X)9.INT 2

FIFE-254.1147 8

MJNT 3

885-27.1147 7

8PE-1 35.1NT 27
8P5-234JNT 14
885-“2.lNT 2
H’E-24.lNT 27
885-1 42.1NT 12
885-288.1817 8

8PE-173JNT 9

885-017M111" 4

8P5-1 35.1NT 24
3173-2351117 8

8PE-1 82.1NT 27
8P5-235JNT 12
885-173.!” 14
885-275.1” 3

885—173.le 12
885-288.1”? 12
8PE-224JNT 2
885-288.1147 24
885-201.!” 14
8PE-307JNT 2
885-201JNT 17
8PE-288JNT 14
8P5-22JNT 2
885-288.1NT 23
8PE-208JNT 15
8P5-235JNT 8

8PE-182JNT 17
8PE-235JNT 2
885-174.INT 27
885-238JNT 7

8PE-174JNT 11
885-275JNT 25
8PE-20JNT 21
8P5-275JNT 21
885-247JNT 18

X8581201JNT 19
X858202JNT 18
X858201.INT 19
X858202JNT 21
X85201JNT 2
X8581402.INT 20
X8581401.|NT 4
X8581302.NT 9
X858101JNT 2
X858402JNT 25
X8581201.NT 14
X858102JNT 19
X8581401.INT 18
X858102JNT 21
X858801JNT 17
X858102JNT 2
X85201 .INT 18
X8581202JNT 18
X858801JNT 11
X8581302JNT 8
X8581401JNT 3
X858102JNT 25
X858101JNT 2
X858902JNT 20
X858101JNT 3
X858402JNT 7
“8401.1” 27
XBEFBCZJNT 14
X85201 .INT 2
X858701JNT 8
X85201 .1NT 12
X8581 502.111" 8
X858702JNT 24
X858102JNT 4
X858401JNT 24
X858702JNT 8
X858701.INT 25
X858702JNT 12
X858702JNI' 28
X8581803JNT 3
X858702.|NT 28
X8581502.INT 12
X8581801JNT 2
X8581801JNT 3
X8581401.|NT 14
X8581501JNT 9
X8581401JNT 17
X8581 5021th 14
X8581501JNT 29
X8581801JNT 2
X8581501.INT 15
X858702JNT 8
X858701JNT 15
X858801.INT 7
X858901JNT 25
X858802.INT 7
X858802JNT 2
X8581701.INT 12
X8581701.INT 21
X85817C1JNT 9
X8589-1JNT 2

11mm
11mm
thuctm
thachlhl
11W
thadnw
XJyachthI
deonb

xnnlonls

Xmolonb

xnnbnb
xaymayzu
Xayzaoayzu
xorynooryzao
Xayzaoayzao
Xayzuoryzu
xaryzaearyno
xayucoryno
layuoayzlcoh
xaynooryzlaah
xayuoovyzboln
xayzaeoryzlcoh
xaynoayzlcoh
Xayzaoayzlcah
Kayneoryzlcah
Xmanicoh
Xmaydcoh
Xphl

X3”

X3“

X.p0pul

Xpopui

xpopul

Xanadu!
111.030an!
)1.“ch
11m

11m

11m
1159731755735:va
xmmnunmm
xmmamm
xmmhmnsanmmm
xnmamm
xmnstnscaulb
11mm
11mm
xmnslwenscaradb'
xmnslucanscomlIB‘
XWSM'
xmwuconsgmmm
xnmsgrunm
xnmlucensgramlnls
xmmconsmlmm
Xhmmcenshordel
X.1rm5lu05nshordol
xnnsluconshordol
xnmlucanshomal
XWSW
xtmnshmnshadal
XWW

”40739‘
”407398
”40740
”407419
”40742
”40241
”40242
”40270
”40271
”40272
”40274
”4050477

”40 5047T8

”4084111
”408518
”4023
”40238
means
”40857
”408878
”40881
”408818
”40885
”40258
”40793
”407938
”40797
”4084711'

”40847118

”408472
”405743
”405753
”40974
”40471
”40478
”404788
”40284
”40285
”4022
”40590
”40591
”407278
”4072711
”407384
”407392
”40887
”40891
”40879
”408798
”4082
”40595
”40713
”40728
”407288

LMG 737(3/7)
LMG 737(9/7)

”408279
”40827911

”4082792

”40882
”40884

8XE-25JNT 8
8XE-241JNT 19
8XE-239JNT 14
8XE-210JNT 8
BXE—OOSJNT 8
8XE-002JNT 7
8XE-27OJNT 5
8XE-207JNT 19
8X5-239JNT 20
8XE—24JNT 8
8XE-24OJNT 2
8XE—004JNT 4
8XE-272.INT 2
8XE-218JNT 1 1
8X5-271 .INT 3
8XE-181 .INT 25
8XE-257JNT 7
8XE-1 82.1NT 1 5
8XE-241JNT 4
8X5-193.1NT 18
8XE-24OJNT 9
8X5-12JNT 17
8XE-013JNT 8
8X5-12JNT 2
8XE-240.INT 19
8XE-193JNT 7
8XE-271JNT 2
8XE-004.INT 7
8XE-270JNT 15
8XE-182.INT 9
8XE-012JNT 7
8XE-23JNT 8
8XE-012JNT 8
8XE-002JNT 9
8XE-237JNT 5
8XE-1 93.1NT 1 5
8XE-005JNT 8
8XE-1 2.lNT 18
BX5-25JNT 14
8XE-193JNT 3
8XE-25JNT 15
8XE-241JNT 2
8XE-(20JNT 4
BXE-1 2.lNT 12
8XE-25.INT 17
8X5-1 39.111” 1 1
8XE-237JNT 20
8XE-004JNT 24
8X5-254JNT 8
8XE-1 2.lNT 1 0
8XE-238JNT 24
8XE-210.INT 7
8XE-25JNT 3
8XE-1 59.1NT 8
8XE-241JNT 10
8XE-1 82.1NT 21
8XE-039JNT 20
8XE-222JNT 5
8XE-280JNT 25
8X5-24JNT 28
8X5-22.INT 28

82

505-255JNT 8
505-244JNT 19
505-231.INT 14
505-212JNT 8
505-018JNT 8
505—01 8.1NT 7
505-273.INT 5
505-208.1NT 19
505-231JNT 20
505-007JNT 10
505-232JNT 2
505-055JNT 3
EOE-272.111" 2
505-21 9.1NT 1 1
505-274JNT 3
505-1 78.1NT 21
505-281.INT 7
505-179.INT 13
505-244JNT 4
505-21.INT 11
505-232JNT 9
505-195JNI' 17
505-015.INT 5
505-178.INT 28
505-232JNT 19
505-1 95.1NT 7
505-274JNT 2
505-007JNT 7
505—273JNT 15
505-1 79.1NT 7
505-014.INT 7
505-02JNT 8
505-01 4.1NT 8
505-01 8.1NT 9
505-29.INT 5
505-281.INT 15
505-018JNT 8
505-28JNT 13
505-255.INT 14
505-195.INT 3
505-255JNT 15
505-244JNT 22
505-015.INT 2
EOE-28.11117 18
505-255JNT 17
505-175.INT 17
505-229JNT 20
505-007JNT 24
505-258JNT 18
505-28JNT 18
505-230.INT 24
505-378JNT 13
505-255JNT 3
505-230JNT 2
505-244JNT 10
505-179.INT 18
505-015.INT 27
505-23.INT 5
505-378JNT 11
505-015.|NT 28
505-382JNT 5

8PE-27JNT 8
885-247.847 19
8PE-25JNT 14
8PE-217JNT 8
8PE-048.INT 8
8PE—047.|NT 9
885-275.1NT 5
885-208.INT 19
8PE-235JNT 20
8PE-01OJNT 8
885-238.1” 2
8PE-010JNT 4
885-272.1147 2
8PE-20JNT 11
885-278.1197 3
885-173JNT 25
8PE-285JNT 7
8PE—174.INT 17
8PE-247JNT 4
8PE-307JNT 5
8PE-238JNT 9
8PE-27JNT 4
885-018.1117 5
885-174.11" 4
333435.137 20
885-307.1147 2
885-278.1” 2
8P5-010.INT 7
885-2754” 15
885-174.1NT 12
8PE-01 7.11117 7
885-009.1147 8
885-017.1197 8
8PE-047.INT 12
8PE-23.INT 5
885-205.11” 15
8PE-047JNT 11
885-135JNT 18
885-254.1197 2
8PE-205.INT 3
885-21le 15
8PE-247.INT 2
885-018JNT 2
8PE-135JNT 12
8PE—27JNT 17
8P5-181JNT 17
8PE-233JNT 20
8PE-010.|NT 24
8P5-254JNT 10
8PE-135JNT 10
885-24JNT 24
885-217JNT 7
885-27JNT 3
885-“2.lNT 8
8P5-247JNT 10
8P5-174.INT 23
885-033JNT 8
8P5-24JNT 5
8P5-308.IN1'2
8?5-010.INT 28
885-282M 28

X858402.NT 8
X8589-1JNT 3
X858702.INT 14
X8581 502.1197 28
X85822JNT 8
X858301JNT 7
X8581803.1NT 5
X8581501JNT 19
X858801JNT 8
X858201.NT 8
X85822JNT 2
X858201.INT 4
X8581802.|NT 2
X8581803JNT 24
X8581702JNT 3
X858801.INT 25
X8581202JNT 7
X858901.|NT 15
X858902JNT 4
X8581201JNT 18
X85822JNT 9
X8581201.1NT 17
258202.111" 5
X858802JNT 18
X858901.|NT 8
X8581201.“ 7
X8581801JNT 9
X858901JNT 9
X858202JNT 7
X8581803JNT 15
X858102JNT 7
X28101JNT 8
X858102.NT 8
X858301JNT 9
)(858502JNT 5
X8581 201.1NT 15
X858302JNT 8
X858401.INT 18
X858402.|NT 14
X8581201 .1NT 3
X8584C2JNT 15
X8589-1JNT 5
X858202JNT 2
X858401JNT 12
X858402JNT 17
X858501JNT 11
X858502JNT 20
X858201JNT 24
X858902JNT 23
X858401JNT 10
X858701.INT 5
X8581502JNT 27
X858402JNT 3
X85821 .INT 8
X858902JNT 10
X858901.INT 21
X858202JNT 27
X8581702.INT 17
X8581801JNT 5
X858201.INT 28
X8581801JNT 8

 

X unamphbl LMG 719 9115-1911!" 19 505-1791NT 15 RPS-173.1!" 20 $5890111" 19
X W pub! LMG 719 9115-2401!“ 11 505-2591!" 13 8PE-23911" 11 $589021!" 11
X W pm! L140 723 9115-1991NT 7 EOE-391.11" 9 8PE-201 .11" 7 $58140111" 7
X. We phlei LMG 730 9115-0201!" 9 EOE-055.11" 9 8P5-0331N1’ 12 $582021NT 29
X. W pou LMG 594 9115-2071911 24 505-2941NT 15 81’5-2131!" 24 $581501 .INT 24
X. 119m p099 LMG 729 9115-00211" 2 EOE-019.11" 2 8P5-0471!" 4 $583021NT 2

X. W scab LMG 7507 9115-21911" 9 505—2191NT 9 8P5-2201!" 9 $5819031!" 19
X W soul‘s L140 993 9115—0451NT 11 EOE-03111" 29 8P5-0471NT 5 $583021!" 3
X um 173nm LMG 5259 9115-19211" 4 505-17911" 2 81’5-1 7419" 9 $589021!" 19
X. “mam mm ”40 529011 9115-2401!" 20 EOE-232.11" 20 8P5-2391NT 21 $589011!" 9
X um ”mam LMG 529012 9115-1911!" 23 EOE-179.1!" 19 8P5-1731!" 23 $58901 .INT 23
X m transmit L140 5292 9115-271 .11" 20 505-2741!" 20 8P5-2791NT 20 $58190111" 9
X um um ”40 975 9115-22219" 11 505-2741!" 23 8P5-2241N1’ 11 $5817021NT 23
X W W L140 979 9XE-00511" 4 EOE-031.1!" 29 8PE—0471NT 9 xaemczmr 4
X m unddaa L140 9293 9115-22211" 20 EOE-223.11" 19 8P5-2241!" 20 $58190111" 20
X. W madam L140 995 9115—24011" 4 EOE-232.11" 4 8PE-2391NT 4 $589021!" 4
X mm» mm L140 999 9115-21911" 4 EOE-219.1!" 4 8P5-22011" 4 11958190311" 19
X W arm L140 999 9115-2701!" 29 505-37911" 20 8PE-30911" 13 $581 70111" 13
X W: m L140 22 9115-01211" 5 EOE-009.1!" 5 8P5-04911" 5 $58301 .INT 5
X. mlcola holdcda LMG 73911 9115-23911" 29 505-2301N‘1' 29 8P5-2341!" 29 $58701 .11" 9
X um holdcoln LMG 73911811 9115-19319" 13 EOE-195.11" 13 8P5-2051!" 13 $58120111" 13
X when holdaoh LMG 7391188 9115-29011" 29 5054921!" 9 81’5-30911" 5 $58190111" 9
X mlcoh m L140 73912 9115-1998" 3 505-1791NT 2 8P5-14211" 3 $585021NT 25
X mica. ham LMG 72128 9115—31511" 19 505-3011!" 29 8PE-30711" 20 $581 40211" 3
X vulcoh holdcola L146 7419 9115-0041!" 9 EOE-007.11" 9 8P5-01011" 9 119582011117 9
X Mach ham LMG 742 9115-23911" 20 EOE-230.1!" 20 8P5-23411" 20 $58701.11" 2

X mlcoh holdcoh L146 7428 9115-1931!" 10 505-1951!" 10 8P5-2051N'1’ 10 $5812011NT 10
X vuicob holdcola LMG m 9x5-2701!" 9 505-27311" 9 8P5-27511" 9 $5819131NT9
X mlcoh holdcdl LMG 92798 9115-22211" 23 EOE-223.11" 22 8PE-2241N'1’ 23 1195819011!" 23
X ndcoh hound. L140 9277 9115-2701NT 13 EOE-273.1!" 13 8P5-2751N‘1’ 13 1195819031!" 13
X munch mcdonm ”40 3234 9115-21911" 15 5115-2191911 15 8PE-2201NT 15 $581921!" 29
X mloah Wm L140 900 9115-241 .11" 3 505-24411" 3 8P5—24711" 3 $589021NT 3
X. vndcoh W ”40 902 9115-1921!" 23 EOE-179.11" 21 8PE-1741!" 25 $58901 .11" 23
X mm LMG 911 9115—0131!" 9 505-0591N‘1' 4 8P5-0191NT 9 $5202.11" 9
X mm LMG 919 9115-191 .INT 27 EOE-179.1!" 23 8P5-1731NT 27 $589011NT 27
X mm LMG 917 9115-2401NT 17 505-2321NT 17 8P5-23911" 19 $589011NT 3
X mm L140 919 9115-2221!" 7 505-2231NT 7 8P5-2241!" 7 $581702le 19
X mm L140 92011 9115-01311" 10 505-0591!" 5 8P5-01911" 1O $582021!" 10
X mic-m LMG 925 9115-2191911 14 505-2191!" 14 8PE-2201NT 14 $58190311" 27
X ”slanted: LMG 935 9115-2711NT 29 EOE-274.1!" 29 8P5-2791NT 29 $581901 .1!" 17
Non-target Data

X abodoola calyilnc' L140 9959 9115-0031!" 12 505-0091!" 12 8PE-0091NT 12 $581011!" 12
X m1: cayilrn' ”413 9959 9115-0121117 12 505-0141NT 12 8P5—0171NT 12 $581021!" 12
X arbalcola popull LMG 12141 9115-003.le 19 505-0091NT 19 8P5-0091!" 19 $581011!" 19
X 09011:»!- papal LMG 12141 9115-0391NT 5 505-0141NT 19 8P5-0171NT 19 $581021NT 19
X axonopodis cassava LMG 9049 9115-2591!" 13 505-301 .1!" 15 895-2001!" 12 $581301JNT 12
X axonopodis came [MG 9049 9115-3051!" 10 505-37911" 5 8P5-2991NT 14 $5813021NT 13
X. axonopodis casslao L146 975 9115-19911" 15 505-2021NT 15 8P5-2001!" 15 1195813011NT 15
X axonopodis m LMG 975 9115-3051!" 13 505-39011" 22 8P5-3071NT 15 $5813021NT 19
X mnpodls elm LMG 9045 9115-1991911 14 EOE-202.1!" 14 8P5-2001NT 14 $58130111" 14
X axonopodis clltodu LMG 9045 9115-25911" 15 505-2921NT 15 8P5-29911" 19 $5813021NT 15
X. axonopods coma LMG 7479 9115-1591NT 9 505-1 79.11" 5 8P5-1421NT 9 $585021!" 29
X axonopodis commune LMG 7479 9115-2371911 29 505-2591NT 9 8P5-2331!" 29 X9589011NT 9
X axonopodis cyamopslds LMG 991 9X5-19911" 9 EOE-202.11" 9 8P5-2001NT 9 $581301 .INT 9
X. axonopods cyamopslds LMG 691 9115-2591!" 7 EOE-29211" 7 RPS-299.11" 9 $58130211" 7
X macpodls dosmodl LMG 992 9115-25911" 5 505-2021NT 4 8P5-2001NT 4 $581301.INT 4
X. wropodis deemed! LMG 992 91(5-3051NT 9 505-3901NT 19 8P5-2661!" 9 $5813021NT 5
X. wrapodls dosmodflgangou LMG 993 9115-19911" 29 EOE-202.1!" 28 8P5-200.I!" 29 $58130111" 29
X axonopodis desmodugangad LMG 993 9115-2591!" 29 505-3611NT 2 8P5-2671NT 2 $5813021NT 29
X axmopods doanodimdno LMG 9049 9x5-1991!" 19 505—2021NT 19 8P5-200.INT 19 $581301 .INT 19
X. axonopodis WW LMG 9049 9115-25911" 20 505-2921NT 20 8P5-2991!" 22 $5813021NT 20

83

.xanmqmubdnmaflmwn!
‘xanquUbdumwdhmmd
xuuawnmkoquNo
.XImwnmwkawmmuo
)(anquflbkmmmhun
JCanmququndumn
)(anquflkpMMwWfi
:XanquukpMMqu
)(anquubpdmwwkdh
)Canquukpdmwwkdh
JCanmqmdbdymmamn
)(anqudkfiym$amu
)(anmqmwksaflnnbo
‘xanmqmumsubmmw
JCAnquUEMQMMHMhna
)(anquflbmfimwnflhflo
:XanmqmdsdwkUb
)Canmnmmkvflmkdb
)(hmmnxgmmhb?
)(bmmhvgmmhb?
:Xbmmhxgmmmn?
:xbmmnxgmmhb?
xuamnMMBMNMMn
xnamanBummun
xuampumbammmnhe
xuampumbamwnwho
.Xcamnumbammmnho'
)Caummnflbamxmuflm'
:XUImMDawhadu‘
)cmwnMuunhaUN'
.xummmmamhwuu”
:xammunuwhauu”

:xammnxamuwmwmnuub
:xnmnumxmspmupMMMnb

LMGG94
LM0694
M898
LMG898
LMG757
LMG757
LMGB44R
LMG844H
LM6849
LMG849
LMG8021
LMGW1

LMG887

LMGM411
LMG827481
LMG827412
LMG827412
LMG9037
moses?
LMG738312
mom
LMG535
LMGS35
LMG72O
LMG720
LMG879
LMGB79
LMGB43
W848

BXE-198JNT 17
BXE-258JNT 18
BXE—21OJNT 14
BXE-289JNT 5
BXE-207JNT 7
8XE-259JNT 27
BXE—210JNT 4
BXE-ZBOJNT 24
BXE-182JNT 18
BXE-241JNT 8
BXE-207JNT 10
BXE-28OJNT 2
BXE—218JNT 17
BXE-271JNT 2
BXE-218JNT 24
BXE-271JNT 8
BXE-222JNT 14
BXE-271JNT 28
BXE-139JNT 19
BXE-237JNT 8
BXE-1 59.1NT 2
BXE-237JNT 24
axe-001 .INT 8
BXE-(XMJNT 12
BXE-(XH .INT 8
BXE—WJNT 14
BXE—001JNT 7
BXE-004JNT 13
BXE~207JNT 8
BXE-259JNT 28
BXE-159JNT 28
BXE-238.INT 19
BXE-OO4JNT 29
BXE-O20JNT 9

ECE-202JNT 17
ECE-282JNT 18
EOE-2121M 14
ECE-378JNT 18
ECE-208JNT 7

EOE-263.11" 27
ECE-381JNT 29
EOE—378JNT 10
ECE-179JNT 14
ECE-244.|NT 8

ECE—208JNT 10
ECE-284JNT 2

ECE~219JNT 17
ECE-378JNT 22
EOE-219.1NT 24
ECE-274JNT 8

ECE-223JNT 13
ECE-274JNT 28
ECE-175JNT 9

EOE-229.18” 8

ECE-175JNT 27
ECE-229JNT 25
EOE-MINT 14
ECE-O15JNT 12
EOE-0151M 14
EOE-081.11“ 18
ECE-015JNT 13
EOE-m1 .INT 17
ECE-m1 .INT 21
ECE—378JNT 8

EOE-178.18” 25
ECE-230JNT 19
EOE-056.1?“ 7

ECE-058.1NT 7

RPS—2881M 19
RPS-$7.1m 18
RPE-217JNT 13
RPS—308JNT 7
RPE—208JNT 7
RPE-287JNT 28
RPE—Z17JNT 4
RPE~268.INT 25
RPE-174JNT 18
RPE—247JNT 8
RPE—2OBJNT 10
RPS—2881M 2
BPE~220JNT 17
RPE—308.INT 14
RPE~220JNT 24
RPE-Z78JNT 8
FIFE—2241M 14
RPE-278.INT 28
RPE—181JNT 9
HPE-233JNT 8
HPE—233JNT 24
RPS-25MB" 3
RPE~01OJNT 12
RPE-O18JNT 12
RPS-010.18" 14
HPE-O18JNT 14
RPE~01OJNT 13
FIFE—0181M 13
RPS-208.1” 8
RPS—287.18“ 29
RPE—182JNT 3
RPEu254JNT 5
FIFE—018.18" 29
RPEo-038.INT 14

XBER1301JNT 17
XBER13C2JNT 18
XBER1801.INT 14
XBER18C2JNT 5
X85814C2JNT 27
XBER1SC1JNT 7
XBER1SC2JNT 24
XBER18C1JNT 4
XBERQCLINT 18
XBER902JNT 8

XBER1SC1JNT 10
XBER1 502.1NT 2
X85317C1JNT 17
mER17C2JNT 2
XBER17C1JNT 24
X85317C2JNT 8
X85317C2JNT 28
”831801.18" 14
XBERSC1 .INT 19
XBERSC2JNT 8

XBERSC2JNT 24
XBER801JNT 2

XBER201JNT 12
“18202.1!" 12
XBER201JNT 14
XBEMCZINT 14
XBER201JNT 13
XBER202JNT 13
XBER14C2JNT 28
XBER1501JNT 8
XBERBCHNT 28
XBERBC2JNT 19
XBER2C1JNT 29
XBER202JNT 29

Figure A.l Sample rep-PCR Genomic Fingerprint Data File (First Two Strains)

 

 

List: NNBOXAPS
Traces: 752

Standard: JANGELOZ 12

Resolution: 400

D: \JANSUB\GELS . INT\BXE--004 . INT 2
Xanthomonas axonopodis vesicatoria

LMG 910
9
39 39 39 39
53 32 16 9
5 2 0 0
16 18 21 25
73 82 85 89
50 41 34 30

5 2 0 O
9 5 5 O

219 255 254 197 137

25

25
80
23

11 O 2 7 9
O 0 O O 0
2 5 2 0 7

25 27 3O 3O 34

89 107 105 91 66

16 14 5 2 0
O 0 0 11 18
O 0 O 14 23

94 105 117 130 128 130 130 133

84

16

5
11
64
71

2
53
39

23 27

5 21
18 14
91 158
73 69

5 11
62 55
43 46

37
25
16
199
59
14
27
46
117

43 46 32
25 11 0
18 25 23
190 137 80
53 62 69
11 9 5
16 21 21
55 66 69
112

41 48

2 2
21 18
73 71
75 64

7 7
18 11
98 153

119 135 139 130

 

 

117
27
80
46
46
55

103

34
187
103

87
30
55
43
41
48

119

27
174
126

71
34
23
59
50
32

151
16
155
151
94

78
41
11
80
57
14

158

137

158
139

107
25

114
71

16
165

105
135
144

126

0
130
82

25
135

71
96
91

112

144
110

21
98

39
62
41

85
9
0

64
37
9

121 114
126 114

0
21
78

9
23
37

0
34
80
21

9
21

43
57
48
137
80

48
107
32

OQO

37
59
85
181
48
0
69
126
62

30
37
133
222
53
5

142

27

139
235
62
25
62
137
155

23

146
249
71
57
55
130
192

23
16
96
210
80
80
62
112
222

21
41
64
137
85
78
85
101
233

27
91
34
73
85
53
105
78
245

30
105
41
48
73
23
110
59
231

D:\JANSUB\GELS.INT\BXE--OO4.INT 3
Xanthomonas axonopodis vesicatoria

LMG 929

9

OWNO

55
30

12
233
67
23
46
32
32
18

106

99
136
106

ONNO

60
30

255

MONO

69
23

214
42
3O
21
46
46
16

145

18
131
155
157

UIONO

69
14

150
74
28

65
48

138

16
115
161
155

UIOU'IO

71

\1

81
101
14

81
65

106
16
85

145
99

\IUINO

71

ON

60
111

69
88

69
18
55
131
55

01000

60

O

69
99
12
23
53
113

OON

14
51

OO

88
71
42
46
37
108

OUIUI

OUIUI

58

28
16
120
28
74
97
127
42

39
104
30
5

9

O

OQW

65

35
21
113
32
44
125
196
30
16
51
111
55
12
0

0

14

157
58

23
21
106
58
14
141
233
39
23
51
104
99
21
2

0

14

191
46

14
28
83
78

122
228
46
48
62
99
150
35
5

0

12

138
32

28
83
85

81
164
53
69
74
99
180
55
14
2

O

88
25

30
95
58
30
39
99
62
74
101
111
180
69
23
21

OOO

48
21
14

42
106
32
58
23
44
55
62
115
138
168
78
32
21

OU‘UI

46
18
14

55
120
23
97
25
39
51
35
122
161
145
81
35
12

12

46
21

106
115
21
85
30
30
28
21
113
175
138

32
114
46
50
59

96
43
208
85

14

44
28

14
166
99
23
74
32
30
23

99
148
129

88

46

 

85

 

APPENDIX B

DATABASE AND DATA FILES

Table 8.1 rep-PCR Genomic Fingerprint Database Properties

 

 

Table Column Type Size

Name

Record Count: 89 Name Text 50

Strain

Record Count: 376 Strain Text 40
Name_ Text 50

BFIngerprlnt

Record Count: 752 Index Text 40
Strain_ Text 20
Fingerprint Memo -

EFIngerprlnt

Record Count: 752 Index Text 40
Strain_ Text 20
Fingerprint Memo -

RFlngerprInt

Record Count: 752 Index Text 40
Strain_ Text 20
Fingerprint Memo -

BCIassItIcatlon

Record Count: 752 Class_ld Number (Integer) 2
Index_ Text 40
Filtered Memo -
Classes Number (Integer) 2
Classifications Text 140
Set Text 10

ECIassIflcatIon

Record Count: 752 Class_ld Number (Integer) 2
Index_ Text 40
Filtered Memo -
Classes Number (Integer) 2

 

86

 

RCIassIflcatlon

Record Count: 752

Target

Record Count:

Non-target

Record Count:

BParameters

Record Count:

EParameters

Record Count:

RParameters

Record Count:

63

26

16

16

16

Classifications
Set

Class_ld
lndex_
Filtered
Classes
Classifications
Set

Tar_ld
Name_
ClassVec

Non_ld
Name_

Parameter
Value

Parameter
Value

Parameter
Value

Text
Text

Number (Integer)
Text

Memo

Number (Integer)
Text

Text

Number (Integer)
Text
Memo

Number (Integer)
Text

Text
Text

Text
Text

Text
Text

140
10

40

140

10

50

15
4O

 

87

Table 8.2 BER Fingerprint Database Properties

 

 

Table Column Type Size

Name

Record Count: 91 Name Text 50

Straln

Record Count: 378 Strain Text 40
Name_ Text 50

BFIngerprInt

Record Count: 752 Index Text 40
Strain_ Text 20
Fingerprint Memo -

BCIasslfIcatlon

Record Count: 752 CIasst Number (Integer) 2
Index_ Text 40
Filtered Memo -
Classes Number (Integer) 2
Classifications Text 140
Set Text 10

Target

Record Count: 63 Tar_ld Number (Integer) 2
Name_ Text 50
ClassVec Memo -

Non-target

Record Count: 26 Non__ld Number (Integer) 2
Name_ Text 50

BParameters

Record Count: 16 Parameter Text 15
Value Text 40

 

88

Sample Data Files:
I Selected Data for Denoising — File Name: *.txo

D:\JANSUB\GELS.INT\BXE--003.INT 16 undex]

LMG 12141 [Strain]

o o o o o o o 4 8 12 8 o o o o o 4

8 4 8 8 8 12 16 12 4 o o o o o o 4 4

8 4 8 12 12 o o o o o 4 8 4 4 o 8 4

o 4 8 12 12 8 o o 4 16 35 35 35 31 31 31 31
1 23 12 4 4 o 4 4 o o o 8 8 19 19 19 4 o
o 12 35 58 66 62 39 23 43 78 105 105 70 35 8 19 23 19
8 4 o o 4 4 8 12 19 12 o o o 4 4 4 4 o
27 101 175 248 213 178 78 35 o 4 12 12 4 31 54 89 7o 54
23 16 8 8 8 4 o o 31 101 171 194 194 190 225 244 255 225
147 66 81 147 217 202 190 128 101 78 93 85 74 43 19 o 4 16
23 27 50 74 101 85 66 27 19 8 54 120 190 217 202 186 116 70
12 8 o o o o o 4 o 16 31 81 93 81 31 4 35 101
167 202 198 178 159 140 101 70 31 27 23 39 39 39 27 31 39 58
85 112 144 159 155 136 120 132 144 128 124 116 105 58 12 o 4 8
4 12 27 so 47 35 19 19 23 19 16 27 50 78 78 47 19 o
8 16 35 31 31 35 62 120 175 221 190 120 35 o 12 19 19 16
12 16 19 23 23 27 35 39 47 62 78 93 124 171 221 240 202 147
85 101 155 213 229 213 194 182 190 186 186 171 136 93 66 81 97 101
74 54 31 35 58 93 112 93 74 35 16 o o 4 16 31 31 19
8 o o o o o o o 4 o o 4 8 16 16 19 27 31
31 27 19 16 16 27 58 85 136 178 202 206 159 116 70 54 66 78
97 101 97 7o 43 16 31 89 144 186 175 163 81 19 o o o o

o o o o 0 [Fingerprint]

8
8
0
3

' Filtered Data — File Name: *.txd

D: \JANSUB\GELS . INT\BXE—-OO3 . INT 16 [Index]

LMG 12141 [Strain]

0 0 0 O 0 1 2 5 6 6 6 4 2 O l 2 4
5 6 7 7 8 10 11 10 9 6 3 1 O 0 1 2 3 5
6 6 8 9 7 6 5 2 O 1 2 3 4 4 5 4 3 2
3 3 5 7 9 8 6 5 6 11 18 25 30 33 33 32 31 29

26 20 15 9 5 3 2 2 2 2 3 7 11 15 14 12 8 7
10 21 34 47 52 50 47 49 58 71 80 79 65 47 31 21 15 15
11 6 3 2 3 6 9 11 10 9 6 3 2 2 3 3 8 27
61 110 153 183 178 150 101 59 26 13 6 13 23 38 50 60 58 50
34 22 13 9 6 4 9 27 61 99 138 170 195 209 222 228 219 187
155 133 132 143 167 177 168 140 118 97 86 75 63 44 28 16 12 14
24 38 55 67 75 71 60 41 35 46 78 118 157 183 182 158 117 78
41 18 4 2 0 1 1 4 10 26 44 60 63 58 49 50 68 102
141 169 181 175 155 130 100 74 50 38 32 33 33 35 35 39 48 65
88 112 131 141 143 140 137 132 130 129 123 106 83 58 36 16 6 6
11 20 28 34 36 34 29 23 19 21 27 38 50 56 54 44 30 18

89

16 18 24 30 39 56 85 123 154 165 148 113 71 37 17 13 16 16
16 17 19 22 25 29 34 42 52 64 81 106 137 170 192 196 179 155
138 140 157 182 201 206 202 193 188 183 174 154 130 109 95 88 84 81
71 59 50 54 66 78 86 81 66 44 25 11 7 10 16 20 21 18
12 5 2 0 0 0 1 1 1 2 3 6 9 13 17 22 25 27
27 25 22 21 27 4O 64 97 132 161 176 172 151 121 93 77 73 79
88 89 82 65 51 50 65 93 125 151 150 125 88 53 20 4 0 0

o 0 o 0 o [Filtered Fingerprint]

I Training Data - File Name: *.txr

D: \JANSUB\GELS . INT\BXE--l3 6 . INT 6 [Index]
10000oooooooooooooooooooooooooooooo

oooooooooooooooooooooooooooorrargetVector]

LMG482[Strain]
00000000000000000

1 2 4 7 10 13 15 17 17 17 17 17 16 17 17 19 21 23
25 25 25 24 22 21 19 17 16 15 15 15 14 13 11 10 8 8
8 7 5 4 2 1 1 2 4 6 9 10 12 13 13 14 14 13
11 9 7 4 2 1 1 1 2 4 5 5 6 5 4 3 2 1
1 1 2 3 4 5 5 6 7 9 10 10 8 7 5 3 2 2
1 1 2 3 5 6 7 7 6 4 3 2 1 1 0 O 1 1
2 2 2 2 1 1 2 3 4 4 4 3 3 3 3 3 2 1
1 0 O 2 4 5 5 5 3 2 1 O 0 0 0 1 1 1
1 1 1 3 12 27 4O 47 45 36 23 12 8 9 9 8 7 5
3 2 1 1 1 2 2 1 2 2 3 4 6 9 11 13 13 13
10 8 5 2 1 0 1 1 1 1 1 1 2 3 5 8 12 18
27 47 77 118 158 192 198 182 142 101 59 34 19 15 13 10 9 8
8 9 9 10 10 10 10 10 10 9 8 7 5 3 1 0 0 0
0 1 3 8 16 27 40 53 66 77 8O 75 4 50 34 23 18 20

27 40 60 88 123 155 173 174 158 130 98 73 57 49 43 38 33 29
24 20 17 15 13 11 9 8 7 6 6 5 4 4 4 5 7 9
11 10 8 5 3 1 1 2 4 7 11 16 24 39 61 91 128 170
197 201 182 151 111 81 69 74 87 104 123 139 144 139 126 107 87 77
73 73 69 62 50 35 20 10 5 2 3 6 11 17 25 31 36 41
45 47 51 51 49 43 37 29 24 21 20 22 25 31 36 38 36 30
22 13 7 3 3 3 3 3 3 1 1 0 0 O 0 0 O 0

o 0 0 0 o [Filtered Fingerprint]

I Testing Data — File Name: *.txs

D:\JANSUB\GELS.INT\BXE--225.INT 22
1 0 0 0 0 0 0 0 0 0 0 O 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 O O O 0 0 0
O 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 O O 0 O 0 0 0 0 O 0 O
LMG 482
0 0 0 O O O 0 0 O 0 0 0 O O 0 O 0
0 1 1 3 5 7 7 7 6 5 5 6 8 8 7 5 5 4

5
29
49
9
5
36
11
14
49
4
6O
51
7
82
17
36
215
48
22
16
0

34

9
38
1
80
56

14
91
16
25
192
51
15

9
0

9 12 14 15 15
24 20 16 12 9
59 61 63 62 60
3 2 1 1 2
5 10 20 32 45
26 17 11 12 19
2 2 2 3 5
7 11 24 44 64
28 17 8 3 1
2 5 9 13 15
101 122 142 159 166
58 55 51 44 38
22 31 39 47 55
106 127 150 170 178
14 16 21 30 40
14 6 1 0 0
147 96 55 36 41
45 35 21 9 1
11 7 4 3 2
3 1 l 1 1
0 0 O

14
7
56

3

51
27
7
78
1

15
164
30

66
171
49
1

61

0
2
2

Non-target Data - File Name: *.txj

12
5
50

5

51
36
7
84
3

13
147
22

80
144
55
1

82

1
2
4

D:\JANSUB\GELS.INT\BXE--003.INT 16

LMG 12141

5
6
3
26
10
11

0
6
6
3
20
21
6

0
7
8
5
15

34
3

0
7
9
7
9

47
2

0
8
7
9

5

0
10
6
8

3

11

6
2

52 50 47

3

6

9

61 110 153 183 178 150 101

34
155
24
41
141

11
16
16
138
71
12
27
88
O

22

133 132 143 167 177 168 140

38
18

13

55
4

9

6

4

9

2
10
2
5
2
49
11

59
27

67 75 71 60 41

2

0

1

1

4

9
0
6
2
58
10

26
61

11 13 18 22 28 31
3 3 3 4 8 15
43 35 28 22 19 18
6 6 5 5 5 5
46 35 24 13 9 9
40 45 47 50 51 48
7 6 7 9 13 17
80 73 67 66 67 69
9 16 24 29 31 28
11 11 12 15 18 21
118 80 48 25 15 15
14 7 3 1 0 O
97 113 124 124 115 102
108 70 40 18 9 8
56 55 52 50 49 51
2 6 13 28 51 87

32 31 30
24 32 41
17 15 12
7 8 7

15 23 33
41 31 20
20 20 18
67 64 57
22 14 8

25 32 43
23 34 43

1 2 3

89 79 77
11 14 17
52 51 45

129 173 206

97 101 96 85 72 56 45 41 44

4 9 17 27 36 41
2 2 3 9 16 23
6 8 8 7 5 3

6 6 6 4 2 0
6 3 1 0 0 1
1 2 3 4 4 5
11 18 25 30 33 33
2 3 7 11 15 14
71 80 79 65 47 31
9 6 3 2 2 3
13 6 13 23 38 50
99 138 170 195 209 222

41 38 30
29 30 23
1 0 0

1 2 4

2 3 5
4 3 2

32 31 29
12 8 7
21 15 15
3 8 27
60 58 50

228 219 187

118 97 86 75 63 44 28 16 12 14
35 46 78 118 157 183 182 158 117 78

10

26 44 60 63 58 49

50 68 102

169 181 175 155 130 100 74 50 38 32 33 33 35 35 39 48 65
88 112 131 141 143 140 137 132 130 129 123 106 83 58 36

20
18
17
140
59

5
25
89
O

28
24
19
157
50

2
22
82
O

34
30
22

182

54
O
21
65
0

36
39
25

201

66
0
27
51
O

34
56
29
206
78
O
40
50

16 6 6

29 23 19 21 27 38 50 56 54 44 30 18
85 123 154 165 148 113 71 37 17 13 16 16
34 42 52 64 81 106 137 170 192 196 179 155
202 193 188 183 174 154 130 109 95 88 84 81
81 66 44 25 11 7 10 16

86

1
64
65

1

1

2 3 6 9 13 17

97 132 161 176 172 151 121 93
93 125 151 150 125 88 53 20

91

20 21 18
22 25 27
77 73 79
4 0 0

I Training Record - File Name: *.txn

Epoch
0
1000
2000
3000
4000

OOOOO

LR

.005
.007
.012
.008
.004

TSSE
1971.

7

3.
2.
1.

001

.682

445
179
584

1971
28
27
26
27

I BPN Output Score - File Name: *.txc

D:\JANSUB\GELS.INT\BXE--225.INT 22

LMG 482
.78
.21
.18
.22
.20
.19

OOOOOO
000000

.19
.23
.20
.20
.20
.21

OOOOOO

.17
.17
.15
.23
.18
.19

00000

.21
.23
.21
.20
.25

00000

.22
.21
.20
.18
.22

.22
.22
.18
.21
.21

00000

GSSE
.209
.687
.042
.956
.119

00000

92

.23
.20
.22
.20
.18

00000

.25
.21
.22
.23
.19

00000

.24
.22
.19
.19
.22

00000

.19
.18
.17
.18
.21

OOOOO

.19
.21
.24
.21
.21

00000

.19
.18
.18
.21
.20

APPENDIX C

PROGRAM LISTING

Mean Filtering (Visual Basic)

Private Sub cmdFilter_Click()
Dim dataPoint(400) As Long, dataFilter(400) As Long
Dim i As Integer, j As Integer, k As Integer

Me.MousePointer = 11

With fMainForm.ProgressBar1
.Visible = True
.Min = 0
.be = txtCount1.Text
.Value = .Min

End With

datafilel = dataPath + '\' + llebName.Caption + 1blType.Caption + '.txo'
FileNuml = FreeFile
Open datafilel For Input As FileNuml

datafilez = dataPath + '\' + llebName.Caption + 1b1Type.Caption + '.txd'
FN2 = FreeFile
Open datafilez For Output As FN2

' Read Data
Do Until EOF(FileNum1)
DoEvents

Line Input #Filethl, Index
Line Input #FileNuml, Header
Line Input #FileNuml, Strain
Line Input #FileNuml, Fingerprint
Line Input #FileNUml, Header

'Mean Filtering
For i = 0 To 399
j = 4 * i + 1
dataPoint(i) = Val(Mid(Fingerprint, j, 4))
Next i

k = ((va1(cboNeighbor.Text) + 1) / 2 - 1)

For i = 0 To k - 1
dataFilter(i) = dataPoint(i)
Next i

For i = 0 To 400 - Val(cboNeighbor.Text)
dataFilter(i + k) = 0
For j = 0 To (val(cboNeighbor.Text) - 1)
dataFilter(i + k) = dataFilter(i + k) + dataPoint(i + j)
Next j
dataFilter(i + k) = dataFilter(i + k) / j
Next i

93

For i = (400 - k) To 399
dataFilter(i) = dataPoint(i)
Next i

Fingerprint = "
For i = 0 To 399
Data = str(dataFilter(i))
If Len(Data) = 2 Then
Data = ' ' + Data
ElseIf Len(Data) = 3 Then
Data = ' ' + Data
End If
Fingerprint = Fingerprint + Data
Next i

’Generate Output
Print #FN2, Index
Print #FNZ, "

Print #FNZ, Strain
Print #FN2, Fingerprint
Print #FN2, "

fMainForm.ProgressBar1.Value = fMeinForm.ProgressBar1.Velue + 1
Loop

Close #FN2, #FileNuml

Call storeParameters('Filtered', 'Mean Filter ' + cboNeighbor.Text)
Me.MousePointer = 0

fMainForm.ProgressBar1.Visible = False

End Sub

Wavelet Filtering (Matlab)
Initialization File: wavinit.m

clear all

addpath D:\PROJECTS\DNAVB
filename = 'berOlB.txo' ;
outputf = ’berOlB.txd' ;
W = ’db8’ ;

level = 3 ;

denoise

Wavelet ToolBox Script File: denoise.m

% wavelet Filtering

fidr = fopen(filename,'r’);
fidw = fopen(outputf,’w');
while 1

% Read data

index = fgetl(fidr);

94

if ~isstr(index), break, end
blank = fgetlIfidr);

strain = fgetl(fidr);

data = fscanf(fidr,'%d');

% wavelet Filtering
datad = wden(data,’minimaxi','s','sln',level,W);
maxdatad = max(datad);

% Nermalized Output
for i = 1:1ength(datad);

if datad(i) < 0

datad(i) = 0;
else
datad(i) = datad(i)/maxdatad * 255;

end;

end;

% Write Filtered data

fprinthfidw,'%s \r\n’.index);
fprinthfidw,'%s \r\n'.blank);
fprint£(fidw,'%s \r\n',strain);

for idx = 1:1ength(datad)
fprintf(fidw,'%3.0£ ',datad(idx));

end;

fprintf(fidw,'\r\n’)3

fprinthfidw.'\r\n');

end;

fcloseIfidr);
fclose(£idw);

fidf = fopen('endflag.txt','w');
fprintf(fidf,'\r\n’);
fclose(£idf);

'End’

Define Ta_rget Classification (Visual Basic)

Private Sub cdeargVec_Click()

Dim SQLl As String, SQL2 As String

Dim vector As String

Dim cell() As Long

Dim N As Long, i As Long, j As Long, count As Long
Dim TrainClass() As String

On Error GoTo ErrHandle
Me.MousePointer = 11
DBGrid1.Visible = False

'Search for unique name

95

SQLB = 'SELECT DISTINCT Strain.Name_ AS BacName FROM BClassification INNER
JOIN (Strain INNER JOIN BFingerprint ON Strain.Strain = BFingerprint.Strain_)ON
BClassification.Index_ = BFingerprint.Index ORDER BY Strain.Name_;'

SQLE = "SELECT DISTINCT Strain.Name_ AS BacName FROM EClassification INNER
JOIN (Strain INNER JOIN EFingerprint ON Strain.Strain = EFingerprint.Strain_)ON
EClassification.Index_ = EFingerprint.Index ORDER BY Strain.Name_;'

SQLR = “SELECT DISTINCT Strain.Name_ AS BacName FROM RC1assi£ication INNER
JOIN (Strain INNER JOIN RFingerprint ON Strain.Strain = RFingerprint.Strain_)ON
RClassification.Index_ = RFingerprint.Index ORDER BY Strain.Name_;'

datUName.DatabaseName = dataPath + '\' + llebName.Caption + '.Mdb'

Select Case lblType.Caption
Case '8'
datUName.RecordSource
Case 'E'
datUName.RecordSource
Case 'R'
datUName.RecordSource
End Select
datUName.Refresh

SQLB

SQLB

SQLR

'Partition into target and non-target classification

SQLB = “SELECT Strain.Name_ AS Name FROM BClassification INNER JOIN (Strain
INNER JOIN RFingerprint ON Strain.Strain = BFingerprint.Strain_)ON
BClassification.Index_ = BFingerprint.Index ORDER BY Strain.Name_;'

SQLE = “SELECT Strain.Name_ AS Name FROM EClassification INNER JOIN (Strain
INNER JOIN EFingerprint 0N Strain.Strain = EFingerprint.Strain_)ON
EClassification.Index_ = EFingerprint.Index ORDER BY Strain.Name_;'

SQLR = “SELECT Strain.Name_ AS Name FROM RClassification INNER JOIN (Strain
INNER JOIN RFingerprint ON Strain.Strain = RFingerprint.Strain_)ON
RClassification.Index_ = RFingerprint.Index ORDER BY Strain.Name_;'

datAName.DatabaseName = datUName.DatabaseName
Select Case lblType.Caption
Case '8'
datAName.RecordSource
Case 'E'
datAName.RecordSource = SQLE
Case 'R'
datAName.RecordSource
End Select
datAName.Refresh

SQLB

SQLR

With fMainForm.ProgressBar1
.Visible = True
.Min = 0
.Max = datUName.Recordset.RecordCount
.Value = .Min
End With

i = 0
ReDim TrainClass(0)
datUName.Recordset.MoveFirst
Do While datUName.Recordset.EOF = False
count = 0
datAName.Recordset.MoveFirst
Do While datAName.Recordset.EOF = False

96

If datUName.RecordsetIBacName = datAName.Recordsethame Then
count = count + 1
End If
If count > 2 Then
ReDim Preserve TrainClass(UBound(TrainClass) + 1)
i=i+1
TrainClasin) = datUName.Recordset!BacName
Exit Do
End If
datAName.Recordset.MoveNext
Loop

datUName.Recordset.MoveNext
fMainForm.ProgressBar1.Value = fMainForm1ProgressBar1.Value + 1
Loop

DoEvents
Call TargetVectorIi, TrainClass)

fMainForm.ProgressBar1.Visible = False
Me.MbusePointer = 0

Exit Sub

ErrHandle:

MsgBox Err.Description
Me.MousePointer = 0
Exit Sub

End Sub

(finerate Target Vector (Visual Basic)

Private Sub TargetVector(N As Long, TrainClass() As String)
Me.MbusePointer = 11

'Clear database

datTar.Re£resh

Do While datTar.Recordset.EOF = False
datTar.Recordset.Delete
datTar.Recordset.MoveNext

Loop

'Unique classification count

txtCount1.Text = N
txtOutputU.Text = N

'Generate target vector

ReDim cell(N)

For i = 1 To N
cell(i) = 0

Next i

i=1
For j = 1 To N
cell(j) = 1
vector = "
For i = 1 To N
vector = vector + str(cell(i))
Next i

97

With datTar
.Recordset.AddNew
.RecordsetlTar_Id = j
.Recordset!Name_ = TrainClass(j)
.RecordsetIC1assVec = vector
.Recordset.Update

End With

cell(j)
Next j

II
0

datTar.Refresh
DBGrid1.Visible = True
DBGrid1.Refresh

Call TrainingTestingSets
Me.MousePointer = 0

End Sub

BPN Training (Matlab)
Initialization File: nninit.m

%% Neural Network Initialization File
clear all

filenamel 'datOlB.txr';
filenamez 'datOlB.txs';
hidden = 120 ;

II = 0.005;

maxepoch = O;

rangel = 31 ;

range2 = 390 ;
matfilename = 'datOlB’;
train_re = 'datOlB.txn’;

BPN Training Script File: nntrain.m

% Neural Networks Training

% Training Data

fidrl = £open(filename1,'r');
i=1;
while 1

% Read data

index = fgetl(fidr1);

if ~isstr(index), break, end

traintarget(:,i) = fscanf(fidr1,'%d’);

strain = fgetl(fidr1);

traindata(:,i) = fscanf(fidr1,’%d');
i=i+1;

98

end
fclose(fidr1);

P
T

2*traindata(range1:range2,:)/255;
traintarget*0.6 + 0.2;

% Testing Data

fier = fopen(filename2,'r’);
i=1;
while 1

% Read data

index = fgetl(fidr2);

if ~isstr(index), break, end

testtarget(:,i) = fscanf(fidr2,'%d');

strain = fgetl(fidr2);

testdata(:,i) = fscanf(fidr2,'%d');
i=i+1;

end

fclose(fidr2);

U = 2*testdata(range1:range2,:)/255;
TU = testtarget*0.6 + 0.2;

% ---- weight Initialization with Nguyen-Widrow initial conditions -—--
R = size(P,1);

L = size(P,2);

81 hidden;

$2 = size(traintarget.1):

p1 = zerosIR,2);
p1(:,2) = ones(R,1)*2;

[W1,Bl] = nwlog(81,p1);

W2 = rands(S2, Sl)*0.01;

82 rands(SZ, 1)*0.01;
% —--- BPN Training ----
disp_freq = 100;
max_epoch = maxepoch;
err_goal = err;

lr_inc = 1.05;

lr_dec = 0.7;

momentum = 0.9;

err_ratio = 1. 4;

TP = [disp_freq max_epoch err_goal lr lr_inc lr_dec momentum err_ratio];
[W1,Bl,W2,BZ,epochs] =
tbpx2(W1,Bl,'logsig',W2,BZ,'logsig',P,T,U,TU,TP,train_re);

% --- saving ---

saveImatfilename,'W1','Bl','W2’,'B2','range1',’range2’)

99

BPN Classification (Matlab)

Initialization File: clinit.m

%% Classification Initialization File
clear all

addpath D: \ PROJECTS \ DNAVB

fnamel = 'datOlB.txs';

fname2 = 'datOlB.txc';

matfilename = ’datOlB';

classify

BPN Classification Script File: classify.m

% Testing Data

% .............
load(mat£i1ename)

fidr = fopen(fname1,'r'):
fidw = fopen(fname2,'w’);
i=1;

while 1

% Read data

index = fgetl(fidr);

if ~isstr(index), break, end
blank = fgetl(fidr);

strain = fget1(fidr);
fingerprint = fscanf(fidr,'%d');

% Classification

finp = 2*fingerprint(range1:range2)/255;
[A1,A2] = simuff(finp,W1,Bl,’logsig',W2,82,'logsig');

result = compet(A2);
fprintf(£idw, '%s \r\n'.index);
fprintf(fidw, '%s \r\n’,strain);
fprintf(fidw, '%3.2f ',A2);
fprint£(fidw, '\r\n'.”);
£printf(fidw, '\r\n’.");

i=i+1;

end

fclose(fidr);
fclose(£idw);

flag = 'ok.’

100

BPN Output Score Thresholding and Pathovar Assignment (Visual Basic)

Private Sub cdeesult_Click()

Dim colNumberNNScore As Long, colNumberCheck As Long

Dim N As Long, Idx As Long, i As Long, j As Long, k As Long, 1 As Long, m As
Long

Dim threshold As Single, score As Single, vflag As Long

On Error GoTo ErrHandle
Me.MousePointer = 11

DataNNinfo.Recordset.FindFirst ('Parameter = 'Output Units")
If Not DataNNinfo.Recordset.NoMatch Then

N = DataNNinfo.Recordset!Value
End If

With DataClassification
.DatabaseName = dataPath + '\' + 1b1DbName.Caption + ".mdb'
.RecordSource = 'Target'
.Refresh

End With

With DataMain

SQLB = "SELECT BFingerprint.Index, Strain.Strain, Strain.Name_ As Name,
RFingerprint.Fingerprint As Fingerprint, BClassification.Filtered As DenFin
FROM BClassification INNER JOIN (Strain INNER JOIN BFingerprint ON
Strain.Strain = BFingerprint.Strain_ ) ON BClassification.Index_ =
BFingerprint.Index ORDER BY BFingerprint.Index;'

SQLE = “SELECT BFingerprint.Index, Strain.Strain, Strain.Name_ As Name,
EFingerprint.Fingerprint As Fingerprint, EClassification.Filtered As DenFin
FROM EClassification INNER JOIN (Strain INNER JOIN EFingerprint ON
Strain.Strain = EFingerprint.Strain_ ) ON EClassification.Index_ =
EFingerprint.Index ORDER BY EFingerprint.Index;'

SQLR = “SELECT RFingerprint.Index, Strain.Strain, Strain.Name__As Name,
RFingerprint.Fingerprint As Fingerprint, RClassification.Filtered As DenFin
FROM RClassification INNER JOIN (Strain INNER JOIN RFingerprint ON
Strain.Strain = RFingerprint.Strain_ ) ON RC1assi£ication.Index_ =
RFingerprint.Index ORDER BY RFingerprint.Index;'

.DatabaseName = dataPath + '\' + llebName.Caption + “.mdb'

Select Case lblType.Caption
Case '3'
.RecordSource = SQLB
Case 'E'
.RecordSource
Case 'R'
.RecordSource
End Select
.Refresh
End With

SQLB

SQLR

colNumberNNScore = fngNScore.Cols
With fngheck
.Visible = False

.Clear

.Cols = 7
colNumberCheck = .Cols
.Rows = 2

101

.ColWidth(O) = 400

.ColWidth(l) = 3200

.TextArray(0 * colNumberCheck + 1) = 'Index '
.ColWidth(2) = 1000

.TextArray(0 * colNumberCheck + 2) = I'Strain '
.ColWidth(3) = 3000

.TextArray(0 * colNumberCheck + 3) = l'Target Classification”
.ColWidth(4) = 3600

.TextArray(O * colNumberCheck + 4) = 'NN Classification'
.ColWidth(S) = 660

.TextArray(0 * colNumberCheck + 5) = "Classes“

.ColWidth(G) = 1000

.TextArray(0 * colNumberCheck + 6) = “verification“
End With

With fMainForm.ProgressBar1
.Visible = True
.Min = 1
.Max = fngNScore.Rows - 1
.Value = .Min

End With
1 = 0
m = 0

For i = 1 To (fngNScore.Rows - 1)
fMainForm.ProgressBar1.Value = 1

Entry = i
Entry = Entry & Chr(9) & fngNScore.TextArray(i * colNumberNNScore + 1)
Entry = Entry & Chr(9) & fngNScore.TextArray(i * colthberNNScore + 2)

'Search for Target Classification

DataMain.Recordset.FindFirst ('Index = " + fngNScore.TextArray(i *

colthberNNScore + 1) + " ')
If Not DataMain.Recordset.NOMatch Then
Entry = Entry & Chr(9) & DataMain.Recordsethame
Else
Entry = Entry & Chr(9) & "
End If

'Determine NN Classification by checking with Target table
threshold = val(txtThreshold.Text)
Entry = Entry & Chr(9) & "

vflag = 0
k = 0
For j = 3 To N + 2
score = Val(fngNScore.TextArray(i * colNumberNNScore + j))
If score >= threshold Then
txtClass.Text = j - 2

DataClassification.Recordset.FindFirst ('Tar_Id = ' + txtClass.Text

If Not DataClassification.Recordset.NbMatch Then
Entry = Entry & DataClassification.RecordsetIName_ _
& ' [' + strtscore) + 'J' & ' '
End If

'Check If NN Classification is Correct

If DataMain.Recordset!Name = DataClassification.Recordsethame_

Then

102

End If

k=k+1
End If
Next j

’Count number of classes
Entry = Entry & Chr(9) & k
If k = 0 Then m = m + 1

' Verification
If vflag = 1 Then
Entry = Entry & Chr(9) & 'T'
Else
Entry = Entry & Chr(9) & " .
End If

fngheck.AddItem Entry
Next i

fngheck.RemoveItem 1
fngheck.Refresh

’Error Analysis

fMainForm.ProgressBar1.Visible = False

subError = (fngheck.Rows - 1 - 1 - m) / (fngheck.Rows - 1) * 100

rejError = m / (fngheck.Rows — 1) * 100

1blRec.Caption = 'Recognized: ' + str(l) + ' / I + str(fngheck.Rows - 1) + ' =
I + Format((l / (fngheck.Rows - 1) * 100), “##0.00') + '%'

1blSub.Caption = 'Misclassified: ' + FormatIsubError, '##0.00') + '%'
lblRej.Caption = 'Rejected: ' + Format(rejError, '##0.00') + '%'

Call storeParameters('Threshold', txtThreshold.Text)
If cboClassData.Text = 'an-target' Then
Call storeParameters('False Accept", Format(subError, '##0.00') + '%')
Else
Call storeParameters('Recognition', Format((l / (fngheck.Rows - 1) * 100),
'##0.00') + '%')
Call storeParameters('Misclassified', Format(subError, '##0.00') + '%')
Call storeParametersI'False Rejection“, Format(rejError, '##0.00') + '%')
End If
Call storeClassification
fngheck.Visible = True

Me.MousePointer = 0
Exit Sub

ErrHandle:
Me.MousePointer = 0
MsgBox Err.Description
Exit Sub

End Sub

103

LIST OF REFERENCES

104

LIST OF REFERENCES

Antonini, M., M. Barlaud, P. Mathieu and I. Daubechies. 1992. Image Coding Using
Wavelet Transform. IEEE Transactions on Image Processing, 1(2): 205-220.

Barclay, V.J. and RF. Bonner. 1997. Application of Wavelet Transforms to Experimental
Spectra: Smoothing, Denoising, and Data Set Compression. Analytical Chemistry,
69(1): 78-90.

Boddy, L., C.W. Morris, M.F. Wilkins, G.A. Tartan and P.H. Burkill. 1994. Neural
Network Analysis of Flow Cytometric Data for 40 Marine Phytoplankton Species.
Cytometry, 15: 283-293.

Castleman, KR. 1996. Digital Image Processing. Englewood Cliffs: Prentice-Hall, Inc.

Cho, 8. 1997. Neural-Network Classifiers for Recognizing Totally Unconstrained
Handwritten Numerals. IEEE Transactions on Neural Networks, 8(1): 43-53.

Civerolo, E. L. and M. J. Hattingh. 1993. Xanthomonas campestris pv. Pruni: cause of
prunus bacterial spot, in Xanthomonas (ed. J. G. Swings and E. L. Civerolo).
London: Chapman & Hall.

Curry, B. and DE. Rumelhan. 1990. Neural Network for Classification of Mass Spectra.
Tetrahedron Computing Methodology, 3: 213-237.

Demuth, H. and M. Beale. 1994. Neural Network Toolbox User’s Guide. Natick: The
MathWorks, Inc.

Donoho, D.L. 1995. De-Noising by Soft-Thresholding. IEEE Transactions on
Information Theory, 41(3): 613-627.

Donoho, D.L. and I.M. Johnstone. 1995. Adapting to Unknown Smoothness via Wavelet
Shrinkage. Journal of the American Statistical Association, 90(432): 1200-1224.

Duveiller, E. and H. Maraite. 1993. Xanthomonas campestris pathovars on cereals: cause
of leaf streak or black chaff diseases, in Xanthomonas (ed. J .G. Swings and EL.
Civerolo). London: Chapman & Hall.

105

 

Edwards, M. and DR. Morse. 1995. The potential for computer-aided identification in
biodiversity research. Trends. Ecol. Evol. 10(4): 153-158.

Freeman, J. A. 1994. Simulating Neural Networks with Mathematica. Reading: Addison-
Wesley Publishing Company, Inc.

Goodfellow, M., G.P. Manfio and J. Chun. 1997. Towards a practical species concept for
cultivable bacteria]. In Species: The units of Biodiversity (ed. M.F. Claridge, H.A
Dawah and MR. Wilson). London: Chapman & Hall.

Hartl, D.L. 1994. Genetics. Boston: Jones and Bartlett Publishers.

Haykin, S. 1994. Neural networks: a comprehensive foundation. New York: Macmillan
College Publishing Company.

Hayward, A.C. 1993. The host of Xanthomonas, in Xanthomonas (ed. J .G. Swings and
EL. Civerolo). London: Chapman & Hall.

Hokawat, S. and K Rudolph. 1993. Xanthomonas campestris pv. Glycines: cause of
bacterial pustule of soybean, in Xanthomonas (ed. J .G. Swings and EL.
Civerolo). London: Chapman & Hall.

Jain, A.K, J. Mao and KM. Mohiuddin. 1996. Artificial Neural Networks: A Tutorial.
IEEE Computer, March, 31-44.

Jain, R., R. Kasturi and B.G. Schunck. 1995. Machine Vision. New York: McGraw-Hill,
Inc.

King, RC. and W.D. Stanfield. 1997. A Dictionary of Genetics. New York: Oxford
University Press.

Leyns, F. 1993. Xanthomonas campestris pv. Graminis: cause of bacterial wilt of forage
grases, in Xanthomonas (ed. J .G. Swings and EL. Civerolo). London: Chapman
& Hall.

Leyns, F., MD. Cleene, J. Swings and JD. Ley. 1984. The Host Range of the Genus
Xanthomonas. The Botanical Review, 50(3): 308-353.

Liang, J. 1997. Highly Scalable Image Coding for Multimedia Applications. Proceedings
ACM Multimedia 97 Seattle Washington, 11-16.

Macdonell, M.T. and RR. Colwell. I985. The contribution of numerical taxonomy to the
systematics of gram-negative bacteria. In Computer-assisted bacterial
systematics. (ed. by M. Goodfellow, D. Jones and F.G. Priest). London: Academic
Press.

106

Margulis, L. and KV. Schwartz. 1998. Five Kingdoms. New York: W. H. Freeman and
Company.

Mew, T.W.1993. Xanthomonas oryzae pathovars on rice: cause of bacterial blight and
bacterial leaf streak in Xanthomonas (ed. J.G. Swings and EL. Civerolo).
London: Chapman & Hall.

Misiti, M., Y. Misiti, G. Oppenheirn and J. Poggi. 1996. Wavelet Toolbox User’s Guide.
Natick: The Math Works, Inc.

Pankhurst, Richard J. 1991. Practical taxonomic computing. Cambridge: Cambridge
University Press.

Priest, F. and B. Austin. 1993. Modern Bacterial Taxonomy. London: Chapman & Hall.

Pruvost, O. and B.Q. Manicorn. 1993. Xanthomonas campestris pv. Mangiferaeindicae:
cause of bacterial black spot of mangoes, in Xanthomonas (ed. J .G. Swings and
EL. Civerolo). London: Chapman & Hall.

Rademaker, J .L.W. and F.J. de Bruijn. 1997. Characterization and Classification of
Microbes by rep-PCR Genomic Fingerprinting and computer-assisted pattern
analysis, in DNA markers: Protocols, Applications and Overviews (ed. Caetano-
Anollés, G., Gresshoff, P.M.). Chapter 10, 151-171. J. Wiley & Sons, Inc.

Rademaker, J .L.W., F.J. Louws, M.H. Schultz, U. Rossbach, L. Vauterin, J. Swings and
F.J. de Bruijn. 1997. Molecular Systematics of Xanthomonas by rep-PCR
Genomic Fingerprinting and Computer-Assisted Pattern Analysis. Poster B6
presented at APS annual meeting.

Rat, 1993. Xanthomonas fragariae: cause of angular leaf spot of strawberry, in
Xanthomonas (ed. J .G. Swings and EL. Civerolo). London: Chapman & Hall.

Ride, M. 1993. Xanthomonas populi: cause of bacterial canker of poplar, in Xanthomonas
(ed. J .G. Swings and EL. Civerolo). London: Chapman & Hall.

Rott, P. 1993. Xanthomonas albilineans: cause of leaf scald of sugar cane, in
Xanthomonas (ed. J.G. Swings and EL. Civerolo). London: Chapman & Hall.

Rumelhart, D.E., B. Widrow and M.A. Lehr. 1994. The Basic Ideas in Neural Network.
Communications of the ACM, 27(3): 87-92.

Schaad, NW. and A. Alvarez. I993 Xanthomonas campestris pv campestris: cause of
black rot of crucifers, in Xanthomonas (ed. J .G. Swings and EL. Civerolo).
London: Chapman & Hall.

107

Schindler, J ., P. Paryzek and J. Farmer III. 1994. Identification of Bacterial by Artificial
Neural Networks. Binary Computing in Microbiology, 6: 191-196.

Schittenkopf, C., G. Deco and W. Brauer. 1997. Two Strategies to Avoid Overfitting in
Feedforward Networks. Neural Networks, 10(3): SOS-516.

Skapura, D.M.1996. Building neural networks. New York: ACM Press.

Stackebrandt, E., R.G.E. Murray and HG. Truper. 1988. Proteobacteria class nov., a
name for the phylogenetic taxon that includes the “Pulple bacterial and their
relatives”. International Journal of systematic bacteriology, 38(3): 321-325.

Stall, RE. 1993. Xanthomonas campestris pv. Vesicatoria: cause of bacterial spot of
tomato and pepper, in Xanthomonas (ed. J .G. Swings and EL. Civerolo). London:
Chapman & Hall.

Stall, RE. and Civero EL. 1993. Xanthomonas campestris pv. Citri: cause of citrus
canker, in Xanthomonas (ed. J .G. Swings and EL. Civerolo). London: Chapman
& Hall.

Strang, G. and T. Nguyen. 1996. Wavelets and Filter Banks. Wellesley: Wellesley-
Cambridge Press.

Stryer, L. 1995. Biochemistry. New York: W.H. Freeman and Company.

Swings, J ., L. Vauterin, and Kersters. 1993. The bacterium Xanthomonas, in
Xanthomonas (ed. J .G. Swings and EL. Civerolo). London: Chapman & Hall.

Towner, K]. and A. Cockayne. 1993. Molecular Methods for Microbial Identification
and Typing. London: Chapman & Hall.

Vauterin, L., B. Hoste, K. Kersters and J. Swings. 1995. Reclassification of
Xanthomonas. Inetmational Journal of systematic bacteriology, 45(3): 472—489.

Vauterin, L., B. Hoste, P. Yang, A. Alvarez, K. Kersters and J. Swings. 1993. Taxonomy
of the genus Xanthomonas, in Xanthomonas (ed. J .G. Swings and EL. Civerolo).
London: Chapman & Hall.

Vidaver, A. K. 1993. Xanthomonas campestris pv. Phaseoli: cause of common bacterial
blight on bean, in Xanthomonas (ed. J .G. Swings and EL. Civerolo). London:
Chapman & Hall.

Vogl, T. P., J. K. Mangis, A. K. Rigler, W. T. Zink and D. L. Alkon. 1988. Accelerating

the convergence of the backpropagation method. Biological Cybernetics, 59: 257-
263.

108

 

Watson, J .D., M. Gilman, J. Witkowski and M. Zoller. 1992. Recombinant DNA.
Scientific American Books.

Weaver, 1.8., Y. Xu, D.M. Healy Jr. and L.D. Cromwell. 1991. Filtering Noise from
Images with Wavelet Transforms. Magnetic Resonance in Medicine, 21: 288-295.

Weeks, P.J.D. and KJ. Gaston. 1997. Image Analysis, Neural Networks and the
Taxonomic Impediment to Biodiversity Studies. Biosiversity and Conservation, 6:
263-274.

Wetherbe, J. and N .P. Vitalari. 1994. Systems Analysis and Design: Traditional, Best
Practices. St. Paul: West Publishing.

Whitten, J .L. and L.D. Bentley. 1998. Systems Analysis and Design Methods. Boston:
Irwin McGraw-Hill.

Wilkins, M.F., L. Boddy, C.W. Mom's and R. Jonker. 1996. A Comparison of Some
Neural and Non-neural methods for identification of phytoplankton from flow
cytometry data. Computer Applications in the Biosciences, 12(1): 9-18.

Wu, C., G. Whitson, J. McLarty, A. Ermongkonchai and T. Chang. 1992. Protein
classification artificial neural system. Protein Science, 1: 667-667.

Wu, C., S. Shivakumar. 1994. Back-propagation neural networks for phylogenetic
classification of ribosomal RNA sequences. Nucleic Acids Research, 22(20):
4291-4299.

Yourdon, E. 1989. Modern Structured Analysis. Englewood Cliffs: Yourdon Press.

Zomorodian, A. and K Rudolph. 1993. Xanthomonas campestris pv. Malvacearum:

cause of bacterial blight of cotton, in Xanthomonas (ed. J .G. Swings and EL.
Civerolo). London: Chapman & Hall.

109

 

"‘l11111111111111Es