"It happened in 1958 that John Kendrew's group determined the three-dimensional structure of myoglobin at a resolution of 6 Å. This first view of a protein fold was a breakthrough at that time. Now, more than half a century later, both experimental and computational techniques have substantially improved as well as our understanding of how proteins and ligands interact. Yet, there are many unanswered questions to be addressed and patterns to be uncovered. One of the most pressing needs in structural biology is the prediction of protein-ligand complexes in aiding inhibitor and drug discovery, ligand design, and studies of catalytic mechanisms. Throughout the past few decades, improvements in computational technologies and insights from experimental data have converged into numerous protein-ligand docking and scoring algorithms. However, these methods are still far from being perfect, and only minimal improvements have been made in the past few years. That might be because current scoring functions regard individual intermolecular interactions as independent events in a binding interface. This thesis addresses existing shortcomings in the conventional view of protein-ligand recognition by characterizing interactions as patterns. Finding that binding rigidifies protein-ligand complexes has led to our design of a robust scoring function that predicts native protein-ligand complexes through the coupling of interactions that rigidifies the protein-ligand interface. Also, the analysis of a non-homologous set of protein-ligand complexes has revealed that binding interfaces are polarized - surprisingly, proteins donate twice as many hydrogen bonds to ligands as they accept, on average, and the opposite is true for ligands. A more in-depth analysis of atom type distributions among H-bond donor and acceptor atoms showed that the discovered trends contain surprisingly strong patterns that are also predictive of native protein-ligand binding. Both the coupling of interactions as well as the distribution of hydrogen bond patterns are currently not captured by other methods and provide new information for the prediction and design of ligands. In the absence of the protein receptor structure, our results show that data from experimental assays can be mined to identify functional group patterns on ligands that are predictive of biological activity. Additionally, we present methods to use functional group patterns to improve the success rate of ligand-based virtual screening. Applied to G protein-coupled receptor inhibitor discovery, this approach has led to the discovery of a potent inhibitor that nullifies the biological response and presents the first instance where virtual screening has been used for aquatic invasive species control. Finally, to overcome current challenges in drug discovery for protein-protein interfaces, a new method for identifying small molecules that block protein-protein interactions is presented. We developed and applied an epitope-based virtual screening workflow to find inhibitors of focal adhesion kinase interactions involved in cancer metastasis. In sum, this work presents both novel insights into the coupling among and trends in intermolecular interactions as well as methods to predict the biological activity of ligands based on patterns of functional groups. Along with the insights gained in this work, computational tools and software for measuring the rigidification that is characteristic of native protein-ligand complexes, analyzing H-bond patterns rigorously, and screening millions of small molecules in hypothesis-driven ligand discovery have been developed and are now being made available to other scientists."--Pages ii-iii.
Read
