DarwinDock and GAG-Dock: Methods and Applications for Small Molecule Docking

Citation

Griffith, Adam Reid (2017) DarwinDock and GAG-Dock: Methods and Applications for Small Molecule Docking. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/Z91Z42GS. https://resolver.caltech.edu/CaltechTHESIS:06122017-230026717

Abstract

Computational modeling is an effective tool in studying complex biological systems. Docking of small molecule ligands in particular is useful both in understanding the functioning of proteins as well as in the development of pharmaceuticals. Together with experiment, modeling can often provide a thorough picture of a given system. Computation can often provide details that are difficult or impossible to determine experimentally, while experiments provide guidance on what calculations are useful or interesting. Our goal is to extend computational modeling, specifically ligand docking, to systems not previously possible, such as the challenging glycosaminoglycan (GAG) systems. In order to do this it was first necessary to develop an automatic way of performing docking without extensive user input and experimental knowledge to narrow the list of candidate poses. DarwinDock represents our efforts in this respect. It is a method for small-molecule docking that separates pose generation and scoring into separate stages, which allows for complete binding site sampling followed by efficient, hierarchical sampling. Our convergence criteria for complete sampling allows for diverse systems to be studied without prior knowledge of how large a set of poses needs to be to span a given binding site, making the procedure more automatic. We also replace bulky, nonpolar residues with alanine, which we refer to as "alanization". This allows the ligand to interact more closely with polar sidechains, which help to orient the ligand. Additionally, alanization reduces the impact of incorrect sidechain placement on ligand placement, a concern that sometimes requires user intervention. With DarwinDock working for standard small molecules, it was then necessary to modify the procedure to work on challenging GAG ligands, which are large and have strong negative charges. A modification to DarwinDock – GAG-Dock – allows the method to be applied to GAGs and protein surface interactions. GAGs are large, linear polysaccharides with strong negative charge. They typically interact with the surfaces of proteins, rather than the cavities favored by most small-molecule drugs. GAG-Dock systematically samples the protein surface for unknown binding sites and modifies the pose generation to allow for large, surface-interacting ligands. GAG-Dock allowed us to study several systems important for neuronal development and answer interesting questions posed by experiment. Finally, we needed a way to validate our predictions for GAG binding sites. We used a systematic approach to identify sets of beneficial mutations to the GAG binding sites by building up from individual in silico mutations. Standard mutation experiments typically employ large mutations, such as arginine to alanine, which decrease or destroy binding. However, such information is not always definitive, as large mutations can have wide-ranging effects beyond direct protein-ligand interactions. Mutations that increase binding, however, are less ambiguous because they must form new interactions with the ligand in order to affect binding energies or affinity. Therefore, we have identified and proposed sets of mutations for our GAG predictions for PTPs, NgR1, NgR3, and EphB3. We encourage our experimentalist colleagues to try these mutations and validate our predictions.

Thesis Files