Development of a high-quality benchmark for scoring function assessment

Abstract

Quantitative computation of drug-target binding affinity has been one of the key problems in molecular-targeted drug design. Scoring functions are fast, robust, and fairly accurate and thus are widely applied in drug design. Not surprisingly, a large number of scoring functions have been developed over the past twenty years. The performance of these methods is certainly not equal. An objective benchmark for assessing scoring functions is thus desired by the users of scoring function for making wise choice among available options. It is also desired by the developers of scoring function for obtaining useful guidance. In recent years, we have established a benchmark, namely Comparative Assessment Scoring Function (CASF), to tackle this critical problem. We proposed a set of indicators for measuring the performance of scoring function and developed evaluation methods accordingly. Very importantly, the scoring process was separated from the conformational sampling process in order to evaluate scoring functions on a common ground. The test set used in CASF, which consisted of non-redundant protein-ligand complexes with high-quality crystal structure and reliable binding data, was carefully selected from the PDBbind-CN database. Compare to the similar benchmarks published by other researchers, our CASF benchmark is more advanced, which has been applied by other researchers since published. We describe the basic methods, evaluation results, and international impact of our CASF benchmark in this review article.