A dataset of 196k protein-ligand docking complexes generated by SMINA, with PLIP protein-ligand interaction analyses

A library of 196k protein-ligand complex with docking result and Protein-Ligand Interaction Profiler that can be used to train generative model, including diffusion models. This is the result of the overall chemical space derived from two resources. First, an initial screening of 50k ZINC compounds was performed and subsequently filtered to 36k compounds. Second, results from a predictive model screening of ChEMBL, 2 million randomly selected ZINC compounds, and DrugBank were generated(overall 156k compounds). This second dataset was used to train an RNN, which was then used to benchmark the RNN-generated compounds against its original training set.