DockBind: A Quality-Controlled Data Set of Modeled Protein–Ligand Complexes Associated with Affinity Data

Abstract Ligand–protein interaction prediction is essential for structure-based drug discovery, but further development of machine-learning scoring functions is limited by a lack of high-quality training data sets. Here, we present DockBind, a curated data set of modeled protein–ligand complexes paired with experimental binding affinity data, generated through a quality-controlled docking workflow. This workflow was developed based on an expanded docking benchmark, where we systematically evaluated free and template-based docking methods as well as ligand similarity and maximum common substructure (MCS)-based pose filtering to generate and identify the most realistic poses. The resulting workflow was then incorporated into a pipeline integrating structural and binding affinity data from Binding MOAD, the Protein Data Bank, ChEMBL, and BindingDB, which resulted in 18,332 complexes across 612 protein targets, with high structural reliability and chemical diversity─the DockBind data set─providing a valuable resource for training and benchmarking machine-learning scoring functions.