Machine learning and molecular simulation-based discovery of novel RIPK1 inhibitors

The Receptor-interacting serine/threonine-protein kinase 1 (RIPK1), a crucial regulator of necroptosis, has been implicated in the pathophysiology of various human conditions, including neurological disorders, inflammation, and cancer. Although RIPK1 is a prime druggable target with several RIPK1 inhibitors in clinical trials, there are no approved medications for therapeutic applications. In this study, we identified potential RIPK1 inhibitors from the Comprehensive Marine Natural Product Database (CMNPD) using an integrated computational strategy. A Gradient Boosting machine-learning model trained on curated ChEMBL data achieved high internal predictive performance (test accuracy = 0.97, F1-score = 0.97, ROC-AUC = 0.99) and was used to screen 5472 marine natural products. Subsequent ADMET filtrations and molecular docking on the RIPK1 kinase domain identified four marine natural products, CMNPD23788, CMNPD14579, CMNPD15831, and CMNPD26709, with high binding affinities (- 10.3 to - 9.9 kcal/mol), comparable to those of existing RIPK1 inhibitors, including the reference inhibitor L8D. Two of the compounds, CMNPD14579 and CMNPD15831, were predicted to be central nervous system penetrant, whereas CMNPD23788 and CMNPD26709 were predicted to have oral bioavailability similar to the reference inhibitor L8D. Furthermore, 200 ns molecular dynamics simulations and MM-PBSA calculations suggested the stability of the complexes formed between CMNPD23788, CMNPD14579, CMNPD15831, and CMNPD26709 and RIPK1. Altogether, the results presented here indicate that CMNPD23788, CMNPD14579, CMNPD15831, and CMNPD26709 are potential RIPK1 inhibitors. Since these findings are based on computational predictions, further investigation, including experimental validation in necroptosis-related neurodegenerative disease models, is required.

The online version contains supplementary material available at 10.1007/s40203-026-00586-8.