To develop new protein-ligand scoring functions with machine learning algorithms, such as Random Forest, has been of significant interest recently. By efficiently utilizing a much larger set of experimental data, Random Forest based scoring functions (RFbScore) can achieve better correlations to protein-ligand binding data with known crystal structures; however, more extensive tests indicate that such enhancement in scoring power often comes with significant under-performance in docking and screening power tests comparing to traditional scoring functions. In our current work, we have carried out a systematic study to understand this paradox: Random Forest based scoring functions’ success in scoring and failure in docking. Based on our novel insights, we have developed a new random forest based feature selection and parameterization framework. Our developed scoring functions, which only employ a small number of important descriptors, can achieve superior performance in all four tests of the CASF-2013 benchmark, including scoring, ranking, docking and screening power tests.

Improving both scoring and docking powers of protein-ligand scoring functions with random forest

251st National Meeting (2016)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.