Exploring structural requirements of HDAC10 inhibitors through comparative machine learning approaches

Abstract

Histone deacetylase (HDAC) inhibitors are in the limelight of anticancer drug development and research. HDAC10 is one of the class-IIb HDACs, responsible for cancer progression. The search for potent and effective HDAC10 selective inhibitors is going on. However, the absence of human HDAC10 crystal/NMR structure hampers the structure-based drug design of HDAC10 inhibitors. Different ligand-based modeling techniques are the only hope to speed up the inhibitor design. In this study, we applied different ligand-based modeling techniques on a diverse set of HDAC10 inhibitors (n = 484). Machine learning (ML) models were developed that could be used to screen unknown compounds as HDAC10 inhibitors from a large chemical database. Moreover, Bayesian classification and Recursive partitioning models were used to identify the structural fingerprints regulating the HDAC10 inhibitory activity. Additionally, a molecular docking study was performed to understand the binding pattern of the identified structural fingerprints towards the active site of HDAC10. Overall, the modeling insight might offer helpful information for medicinal chemists to design and develop efficient HDAC10 inhibitors.

Keywords: Bayesian classification; Decision tree; Histone deacetylase 10 (HDAC10) inhibitors; Machine learning; Molecular docking; Recursive partitioning.