Application and Development of Computational Tools in Drug Discovery
Balius, Trent Erik
The Graduate School, Stony Brook University: Stony Brook, NY.
In this dissertation, I will discuss several interconnected projects motivated by drug development. These projects employ computational techniques to study molecular recognition of a ligand (drug) by a receptor (protein) characterized through use of structural and energetic analysis. In Chapter 1, an introduction to computational drug discovery is presented, and methods used here are discussed. Epidermal Growth Factor Receptor (EGFR) is an important drug target for the treatment of cancer. In Chapter 2, we performed all atom molecular dynamics simulations of clinically relevant mutations of EGFR complexed with erlotinib (Tarceva) and other inhibitors. The per-residue decomposition of intermolecular van der Waals and electrostatic energies - termed here molecular footprints - are useful in characterizing mechanisms of drug resistance. For instance, the resistance to erlotinib and other inhibitors observed for the T790M mutation does not employ a steric clash mechanism as was discussed in the literature. In fact, our results show that favorable van der Waals interactions are increased at this position. Notably, water-mediated interactions were revealed to be highly important for explaining the resistance profiles. Footprints are useful in understanding binding. We observed that a molecular footprint can be computed for any pose including conformers generated by docking. Thus, we developed a footprint-based rescoring function in DOCK 6.5, termed footprint similarity (FPS) score. The FPS scoring method is discussed in Chapter 3 and 4. This tool uses comparison methods (Euclidean distance and Pearson correlation) to quantify footprint similarities between a reference molecule and docked molecules. The FPS score enables users to rank-order virtual screening results where the top scoring molecules have similar interactions as those of a reference. References may include known drugs, natural substrates, and low energy transition states, among other possibilities. This method was validated using pose reproduction, cross-docking, and enrichment studies. In addition, experimental collaborators have identify promising lead compounds from our virtual screening projects (using FPS re-ranking) including those targeting BotNT/A and HIV-gp41. The FPS rescoring method was generalized to use grids enabling footprint-guided docking. This grid-based FPS scoring method has been validated using pose reproduction experiments. Future directions and ongoing projects are also discussed including de novo design. In Chapter 5, we conclude with a description of ongoing projects and ideas for future directions. Finally, we discuss two collaborative projects which are presented in the appendixes. The role of point mutations in resistance of the HIV fusion protein glycoprotein 41 (HIVgp41) to the binding of T20, an FDA approved therapeutic was characterized. Energetic error analysis and membrane contributions were also examined. For HIVgp41 application studies, molecular footprints were shown to be useful in understanding drug-target molecular recognition as well as drug resistance mutations. Motivated in part by participating in a docking symposium entitled "Docking and Scoring: A Review of Docking Programs" held at the American Chemical Society meeting (spring, 2011), DOCK 6.6 was evaluated as an enrichment tool using Receiver operating characteristic (ROC) curves and area under the curve (AUC) analysis on validation databases DUD and Wombat. Enrichment studies demonstrate above random global enrichment, as well as good early enrichments, revealing DOCK as a useful tool for virtual screening applications. All of the projects discussed here demonstrate the strength of computational techniques in drug discovery.