Machine learning-based design, screening, and activity validation of topoisomerase I inhibitors

Abstract

Topoisomerase I (TOP I) plays a vital role in maintaining genomic stability and regulating cellular proliferation. Its overexpression in aggressive cancers such as lung, pancreatic, and breast malignancies highlights its value as a therapeutic target. However, the current TOP I inhibitors face limitations including poor hydrolytic stability, significant toxicity, and the emergence of drug resistance. To address these issues, this study developed a comprehensive QSAR framework that goes beyond traditional methods restricted by limited descriptors or single algorithms. A dataset of 550 high-activity compounds from ChEMBL, BindingDB, and Topscience was systematically screened to build thirty QSAR models combining five molecular fingerprint types with six advanced machine learning algorithms. An optimized artificial neural network model was then employed to rationally design 5938 candidate inhibitors using the sequential attachment-based fragment embedding (SAFE) methodology. These candidates underwent rigorous evaluation through activity prediction, drug-likeness assessment, and ADMET profiling, resulting in seven promising compounds. Among them, three were experimentally validated by MTT cytotoxicity assays, while four novel compounds were further characterized by molecular docking and molecular dynamics simulations. This integrative approach provides a robust theoretical foundation for the rational design and optimization of TOP I inhibitors, facilitating the development of targeted therapies against TOP I-associated cancers.