Machine Learning for Anticancer Activity Prediction of Transition Metal Complexes

While metallodrugs have been known for decades, developing new effective formulations still remains a challenge, underscoring the need for new tools to assist identification of potent metal-containing anticancer agents. In this work, we developed a straightforward data-driven approach to predict cytotoxicity of metal complexes based entirely on their (metal + ligands) composition. To this end, we have manually curated MetalCytoToxDB - a comprehensive experimental database comprising 26500 IC50 values for 7050 Ru, Ir, Rh, Re and Os complexes against 754 cell lines from 1921 articles. Based on these, machine learning models were created to accurately assign the cytotoxicity class within Ruthenium and Iridium subsets, resulting in ROC-AUC = 0.81 and 0.73 for binary classification on cross-validation. Moreover, external validation of the best-performing model on the unseen data was carried out to illustrate its capability to assist the potential anticancer drug exploration process, leading to >2x increase in the hit rate compared to random sampling. The possibility of multi-metal predictions was explored, enabling assessment of cytotoxicity among the complexes of metals, for which experimental data is relatively scarce. The interpretability and limitations of the developed models are discussed, providing data-driven chemical insights into metal complexes cytotoxicity and laying foundation for future work. Finally, a pipeline for effective high-throughput computational screening of ruthenium complexes is proposed. The MetalCytoToxDB is available online for AI-assisted exploration at https://biometaldb.streamlit.app/.