A Consistent Scheme for Gradient-Based Optimization of Protein–Ligand Poses

• DOI: 10.1021/acs.jcim.0c01095
• Publication Date: 2020-12-01T13:17:23Z
• Authors: Florian Flachsenberg; Agnes Meyder; Kai Sommer; Patrick Penner; Matthias Rarey
• Keywords: Broyden–Fletcher–Goldfarb–Shanno algorithm; Benchmark (surveying); Optimization algorithm

Scoring and numerical optimization of protein–ligand poses is an integral part docking tools. Although many scoring functions exist, them are not continuously differentiable they rarely explicitly analyzed with respect to their behavior. Here, we present a consistent scheme for pose gradient-based optimization. It consists novel variant the BFGS algorithm enabling step-length control, named LSL-BFGS (limited step length BFGS), empirical JAMDA function designed prediction good optimizability. The shows high performance in CASF-2016 power benchmark, top-ranking RMSD ≤2 Å about 89% cases. combination significantly higher locality (i.e., no excessive movement poses) than classical while retaining characteristically low number evaluations. freely available noncommercial use academic research.