QSAR modeling to design selective histone deacetylase 8 (HDAC8) inhibitors

HDAC8 inhibitors have become an attractive treatment for cancer. This study aimed to facilitate the identification of potential chemical scaffolds for the selective inhibition of histone deacetylase 8 (HDAC8) using in silico approaches. Non-linear QSAR classification and regression models of HDAC8 inhibitors were developed with support vector machine. Mean impact value-based sequential forward feature selection and grid search strategy were used for molecular descriptor selection and parameter optimization, respectively. The generated QSAR models were validated by leave-one-out cross validation and an external test set. The best QSAR classification model yielded 84 % of accuracy on the external test prediction and Matthews correlation coefficient is 0.69. The best QSAR regression model showed low root-mean-square error (0.63) and high squared correlation coefficient (0.53) for the test set. The validated QSAR models together with various drug-like properties, molecular docking and molecular dynamics simulation were sequentially used as a multi-step query in chemical database virtual screening. Finally, two hit compounds were discovered as new structural scaffolds which can be used for further in vitro and in vivo activity analyses. The strategy used in this study could be a promising computational strategy which can be utilized for other target drug design.