Random Forests Regression and Rescoring Strategy for Identifying Inhibitors of Ubiquitin Specific Protease-7.

Ubiquitin-specific protease-7 (USP7) is a potential target for cancer therapy. In this work, the evaluation and design of USP7 inhibitors are conducted employing a comprehensive computational approach. Initially, a large database of compounds undergoes preliminary screening via similarity and pharmacophore analyses. Subsequently, a secondary selection process is executed based on both quantitative structure–activity relationship modeling and molecular docking. The final selection is refined through rescoring treatments via consensus scores and MM/GBSA binding free energy. A total of 138 experimental USP7 inhibitors were collected for the construction of random forest regression models using four different feature selection methods. 17 top experimentally active inhibitors were used as models in similarity searching, and eight representative USP7-ligand binding models were used in pharmacophore screening, performed over a database of 14 million compounds. The rescoring approach applied in this study offers an alternative and beneficial method for the ranking and selection of desired molecules. Based on the molecular scaffolds of the highly active inhibitors, some new derivatives have been designed and received high predicted scores for inhibition. The inhibitors against USP7 were evaluated using an integrated approach including QSAR, pharmacophore, molecular docking, and binding free energy calculations. Random Forests regression models were well built via several methods of feature selection. The rescoring treatment can be used as a helpful strategy for virtual database screening in drug design. [ABSTRACT FROM AUTHOR]