"One of the longstanding challenges in biology is to connect the wealth of genome data to the phenotypes it encodes. In plants, phenotypes can encompass a variety of traits, but one of interest is that of specialized metabolism, or the production of compounds unique to only specific lineages of plants rather than all plants. This interest comes from the wide range uses of specialized metabolites from defending a crop plant against insect herbivory to using the plant compound as a base for pharmaceuticals. The challenge faced is that due to their high diversity, among other reasons, the genes underlying the process of making specialized metabolites are not well characterized. In addition, specialized metabolism pathways can be induced by stress, however it is unclear how most SM pathways and stress responsive genes are regulated. Therefore, the research in this dissertation focuses on 1) How to identify the function of genes as being involved specialized metabolism 2) What characteristics are shared among specialized metabolism genes (SM genes) 3) How are SM genes and genes involved in response to stress regulated?For the first two questions, chapters 2 and 3 use machine learning modeling to predict specialized metabolism (SM) genes in Arabidopsis thaliana and Solanum lycopersicum (tomato). A shared set of characteristics emerges as being important that includes expression features under biotic stress and in specific tissue types. Additionally, evolutionary and duplication characteristics were important where SM genes tend to be recently and tandemly duplicated, as well as less conserved than genes not in SM pathways. Using these characteristics to build a machine learning model, 85.6% of SM genes in A. thaliana and 76.6% of SM genes in tomato were correctly predicted. Additionally, we show that the superior annotation in A. thaliana is able to make cross-species predictions in tomato as well as improve SM gene predictions relative to the model based only on tomato annotation. The improved model predicts 92.4% of SM genes in tomato correctly. Finally, machine learning is used to predict SM genes to a specific pathway.For the third question, chapter 4 uses machine learning to predict how response to wounding stress is regulated and what regulatory elements are important for an SM pathway that is activated by stress. Important putative cis-regulatory elements were identified for genes differentially expressed under wounding stress and temporal patterns of regulation were discovered. Using machine learning, these putative cis-regulatory elements were found to be important in driving differential expression of genes at different time points after wounding. Additionally, regulatory elements were mapped to the genes in the SM pathway glucosinolate biosynthesis from tryptophan to determine element important for the regulation of this pathway under wounding stress. In this dissertation I examine computational approaches to identify gene function and regulatory mechanisms, highlighting the fact that machine learning can be a powerful tool to make challenging predictions."--Pages ii-iii.
Read
