Abstract
Three techniques intertwining and integrating quantum mechanics (QM), molecular mechanics (MM), and unsupervised machine learning were utilized in the prediction of the toluene-water partition coefficient (logP tol/w) for sixteen drug molecules as part of the ninth iteration of the Statistical Assessment of the Modeling of Proteins and Ligands (SAMPL) series of blind prediction challenges. The three blind submissions yielded mean unsigned errors (MUE) ranging from 1.53-2.93 logPtol/w units. Out of all submissions (ranked and unranked), one of these methods yielded the third lowest MUE of 1.53 indicating an overall increase in errors with respect to predicting octanol-water partition coefficients (logPo/w¬) for similar drug-like molecules. After applying numerous QM and MM approaches into multiscale and data-driven approaches to supplement the initial analysis, MUEs were reduced to 1.00 logPtol/w units when using density functional theory (DFT) on a single conformation, while generating an ensemble of rotamer structures elucidates subtle electronic and structural effects that are not considered in the analysis of a single conformation. Computational approaches developed for these SAMPL challenges will continue to serve as standard predictive tools for rational drug design.
Supplementary materials
Title
Electronic Supporting Information (ESI)
Description
ESI contains information on the online resources used, details on the DFT calibration, the 25-point λ path, and analysis of the ensemble-based methods.
Actions