Overcoming cisplatin resistance with AI-driven metallodrug discovery

Cisplatin resistance poses a significant challenge in cancer treatment, particularly when no alternative broad-spectrum drugs are available. While artificial intelligence has emerged as a powerful tool for drug discovery in organic chemistry, its application to metallodrugs remains largely underdeveloped. We hereby report a deep learning model for the screening and de novo design of platinum anticancer drugs. This model integrates a curated dataset of 3,725 platinum complexes with chemoinformatic representations supporting metal-ligand bonds and charges. After screening a large library of 214,373 known complexes, we selected 19 structurally diverse platinum complexes predicted by the AI model to be either active or inactive in different cancer cell lines. The experimental synthesis and testing of these complexes showed up to 72% agreement with the predictions. Leveraging AI-driven metal-ligand combinatorics for de novo design, we developed the PlatinAI complex, which overcame cisplatin resistance in both in vitro and in vivo experiments through a protein signaling network distinct from DNA damage. Additionally, we integrated our manually curated comprehensive platinum database of 17,732 IC50 values with the metal-supporting deep learning model to create MB Finder (www.mb-finder.org), a free public platform that enables users to search for platinum complexes with reported cytotoxicity and predict the anticancer activity of their desired molecules. This open-access resource provides a foundational tool for next-generation platinum-based drug discovery.