UCSF

There are a plethora of computational tools available in the field of structure-based drug design that attempt to predict the complementarity and binding energetics of a ligand to a drug target of interest. These computational tools often comprise large errors and predict incorrect ligand geometries or select non-binding molecules over true ligands. While these false-postive hits, which are referred to as "decoys", are frustrating, they potentially provide important tests for computational methods. In order to test the utility of these computational tools as well as suggest areas for improvement, I provide a database of decoys from several model binding sites which range in complexity from small cavities to real drug targets. Especially in the cavity sites, which are very simple, these decoys highlight particular weaknesses in sampling and scoring procedures.

With decoys from model binding sites, I examine molecular docking methods, which are used to screen large compound databases to find drug leads. Many of the decoy molecules that make up the high failure rate of docking screens are informative, arguably more so than successful predictions. Second, I examine MM-GBSA rescoring of docking hit lists. These more physically realistic methods have improved models for solvation, electrostatics, and conformational change than do most docking programs. MM-GBSA rescoring with binding site minimization improved the separation of known ligands from known decoys for each of the cavity sites and rescued docking false negatives, but also introduced several new decoys into the top-ranking molecules. Finally, I examine alchemical free energy calculations to predict accurate binding free energies of ligands to the simplest hydrophobic cavity site. This method computed absolute binding free energies with an RMS of 1.9 kcal/mol for a set of known ligands, correctly discriminated between several true ligands and decoys in a set of putative binders, and calculated binding free energies of these with an RMS error of just 0.6 kcal/mol. I consider the origins of the successes and some of the particular sources of failures in docking and rescoring in the model sites as well as the implications for biologically relevant targets.