Comparative accuracy analysis of molecular docking scoring functions using the CASF-2016 benchmark dataset

In this paper, a comparative analysis of the performance of AutoDock Vina, NNScore2, RF-Score-4, CENsible, HGScore, OnionNet-2, PIGNet2, and PLANET scoring functions designed to predict the binding affinity of small molecules to a target protein based on molecular docking data was made using the CASF-2016 benchmark dataset. The study has shown that the deep learning scoring functions PLANET and OnionNet-2 demonstrate the highest accuracy, effectively predicting the affinity of the ligand to the molecular target and increasing the reliability of identifying compound candidates with high potential for inhibitory activity. The obtained data indicate that both PLANET and OnionNet-2 can be used in computational protocols of molecular docking for subsequent calculation of the exponential consensus rank for each ligand and reliable selection of the most probable inhibitors of a given therapeutic target.