Does Accurate Scoring of Ligands against Protein Targets Mean Accurate Ranking?

Abstract

Accurately predicting the binding affinities of large sets of protein-ligand complexes efficiently is a key challenge in computational biomolecular science, with applications in drug discovery, chemical biology, and structural biology. Since a scoring function (SF) is used to score, rank, and identify potential drug leads, the fidelity with which it predicts the affinity of a ligand candidate for a protein’s binding site has a significant bearing on the accuracy of virtual screening. Despite intense efforts in developing conventional SFs, which are either force-field based, knowledge-based, or empirical, their limited scoring and ranking accuracies have been a major roadblock toward cost-effective drug discovery. Therefore, in this work, we examine a range of SFs employing different machine-learning (ML) approaches in conjunction with a variety of physicochemical and geometrical features characterizing protein-ligand complexes. We compare the scoring and ranking accuracies of these ML SFs as well as those of conventional SFs in the context of the diverse test sets of the 2007 and 2010 PDBbind benchmarks. We also investigate the influence of the size of the training dataset and the number of features used on scoring and ranking accuracies. We find that the best performing ML SF has a scoring power of 0.807 in terms of Pearson correlation coefficient between predicted and measured binding affinities compared to 0.644 achieved by a state-of-the-art conventional SF. Despite this substantial improvement (25%) in binding affinity prediction, the ranking power improvement is only 6% from a success rate of 58.5% achieved by the best conventional SF to 62.2% obtained by the best ML approach when ligands were ranked for 65 unique proteins.