Leveraging a Large-Scale Experimental Database of Ruthenium Complexes to Predict Cytotoxicity for Anticancer Purposes Using Machine Learning

While metallodrugs have been known for more than 60 years, 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 simple and fast data-driven approach to predict cytotoxicity of ruthenium complexes based entirely on their ligands composition. To this end, we have manually curated a comprehensive experimental dataset comprising 12439 individual cytotoxicity values for 3247 ruthenium complexes from 921 literature sources. Based on it, machine learning models were created to accurately predict cytotoxicity values and assign cytotoxicity class among diverse organoruthenium compounds, resulting in ROC-AUC = 0.76 for binary classification. Moreover, external validation of the best-performing model was carried out to illustrate its capability to assist the drug exploration process, leading to a threefold increase in the hit rate compared to random sampling. Finally, a pipeline for effective high-throughput computational screening of ruthenium complexes was proposed. The cytotoxicity prediction and the RuCytoToxDB exploration are integrated in a digital tool available at https://rucytotoxdb.streamlit.app/.