Machine learning methods used during the D3R protein-ligand docking Grand Challenges

Abstract

The D3R Grand Challenges (GCs) were a series of prospective blinded protein-ligand docking competitions which attracted community-wide participation. The goal of blind challenges is to benchmark existing tools without the inherent bias factor that may be accompanied when conducting retrospective benchmarks. In the D3R GCs participants were asked to predict poses and affinities of small molecules binding to a range of pharmaceutically relevant protein targets. In recent years the explosive use of machine learning afforded state-of-the-art performances in many domains including applications within the biomolecular sciences. This sparked significant interest to apply novel machine learning methods in docking. In this review we highlight machine learning strategies employed during the D3R docking competitions