3D Structure-based Generative Small Molecule Drug Design: Are We There Yet?

Abstract

Structure-based drug design (SBDD) plays a crucial role in preclinical drug discovery. Recently developed structure-based 3D generative algorithms have the potential to streamline the SBDD process by enabling the generation of novel, druglike molecules based on the structure of a target protein's binding pocket. However, no effective metric currently exists to assess the chemical plausibility of molecules generated by these algorithms, which hampers further advancements in their development. In this study, we propose two novel metrics for evaluating the chemical plausibility of generated molecules. Combined with additional analyses, we demonstrate that existing algorithms could generate chemically implausible structures with some property distributions that deviate from those of known drug-like molecules. The metrics and analytical methods presented here offer model developers and user valuable tools to quantify the chemical plausibility of molecules they generate, ultimately facilitating the application of these algorithms in drug discovery.