Utility of scoring function customization in docking-based virtual screening approaches

摘要

Virtual screening of large chemical libraries plays a key role in lead identification and optimization in rational approaches to pharmaceutical drug discovery. Both ligand-based (e.g. 3D-QSAR) and structure-based (e.g. automated docking) methods are extensively used in such screening experiments. While the former techniques lead to good local models that rationalize the structure activity relations (SAR) in a given series, they provide limited insights into receptor-ligand interactions. Structure-based methods depend on scoring functions that are typically not functional-ized to reproduce local SAR while doing a good job of producing accurate poses and enriching actives seeded in a decoy set. In this study, we have attempted to employ the ability tocustomize the scoring function associated with Surflex-dock technology to develop HIV-1 protease inhibitor-specific scoring functions using a well-defined training set of cyclic urea compounds. The study highlights the significance of various docking features such as protein flexibility, fragment-based core constraints and sampling of the docking space in optimization of docking scores. Applying customized scoring functions improved correlation between experimental and computed docking scores and thusenabled better rationalization of SAR. In addition, the tuned scoring functions show utility in recovery of actives in enrichment studies. Such studies lend themselves to identification of novel ligands as potential HIV-1 inhibitors from the pool of chemical libraries whose activities against HIV-1 protease are unknown.