Chemoinformatics study of HDAC3 and HDAC6 inhibitors

Abstract

Histone deacetylases (HDACs) are essential epigenetic regulators that modulate gene expression, cell proliferation, and differ entiation, and their dysregulation has been implicated in the development and progression of numerous cancers. Among the HDAC family, HDAC3 and HDAC6 have emerged as particularly attractive therapeutic targets owing to their distinct biological functions and disease relevance. In this study, an integrated computational framework combining machine learning (ML) within Python lan guage, molecular docking using GOLD 2022.3.0 package, and quantum chemical calculations utilized by Gaussian 09 was employed to identify the structural determinants governing inhibitory activity toward HDAC3 and HDAC6. In this research, the molecules which are part of ChEMBL database were examined. The developed ML models demonstrated excellent predictive performance, achieving accuracy, precision, recall, and F1-score values approaching 1.0 for HDAC3 and approximately 0.98 for HDAC6, indicating high robustness and predictive reliability. It was found that metrics' values do not change significantly for cross-validation and when val ues of hyperparameters were changes (less then 0.05). Feature importance analysis, supported by molecular docking and quantum chemical simulations, revealed distinct structure-activity relationships between the two isoforms. Smaller electron-withdrawing substituents located near the Zn²+-binding region were found to enhance HDAC3 inhibition, whereas bulkier hydrophobic groups favored HDAC6 binding, providing valuable insights into isoform-selective inhibition. These findings demonstrate the potential of integrated computational approaches to guide the rational design of selective HDAC inhibitors for future epigenetic therapies.