Two- and Three-Dimensional Molecular Representations in Ligand-Based Approaches

Molecular representation plays a central role in ligand-based virtual screening of small molecules for the early enrichment of bioactive compounds. Extended connectivity fingerprints, which are derived from structural formulas, have remained a de facto standard in similarity searching and machine learning-based virtual screening. In contrast, three-dimensional molecular representations provide information captured only from the three-dimensional coordinates of molecules. Two-dimensional molecular representations may not accurately describe protein–ligand intermolecular interactions because molecules exist in three-dimensional space. In this chapter, we discuss situations where three-dimensional molecular representations have benefits over two-dimensional representations and vice versa in virtual screening. A similarity profile, which is a set of similarity values against a reference molecule set, was adopted as a coordinate-independent three-dimensional molecular representation. We first discuss the importance of conformation via similarity searching and its derivatives. The relative merits of three-dimensional representation over extended connectivity fingerprints are then clarified via carefully controlled retrospective virtual screening trials. Furthermore, the three-dimensional similarity profile was compared with other two-dimensional representations in potency prediction.