Targeting Cathepsin L in Cancer Management: Leveraging Machine Learning, Structure-Based Virtual Screening, and Molecular Dynamics Studies

Abstract

Cathepsin L (CTSL) expression is dysregulated in a variety of cancers. Extensive empirical evidence indicates their direct participation in cancer growth, angiogenic processes, metastatic dissemination, and the development of treatment resistance. Currently, no natural CTSL inhibitors are approved for clinical use. Consequently, the development of novel CTSL inhibition strategies is an urgent necessity. In this study, a combined machine learning (ML) and structure-based virtual screening strategy was employed to identify potential natural CTSL inhibitors. The random forest ML model was trained on IC50 values. The accuracy of the trained model was over 90%. Furthermore, we used this ML model to screen the Biopurify and Targetmol natural compound libraries, yielding 149 hits with prediction scores >0.6. These hits were subsequently selected for virtual screening using a structure-based approach, yielding 13 hits with higher binding affinity compared to the positive control (AZ12878478). Two of these hits, ZINC4097985 and ZINC4098355, have been shown to strongly bind CTSL proteins. In addition to drug-like properties, both compounds demonstrated high affinity, ligand efficiency, and specificity for the CTSL binding pocket. Furthermore, in molecular dynamics simulations spanning 200 ns, these compounds formed stable protein-ligand complexes. ZINC4097985 and ZINC4098355 can be considered promising candidates for CTSL inhibition after experimental validation, with the potential to provide therapeutic benefits in cancer management.