Russian chemists have created MetalCytoToxDB, the world’s largest database of metal cytotoxicity, designed to solve a key problem in developing metal-containing anti-tumour drugs: the lack of systematic data, GxP News writes. Unlike existing alternatives, the new platform contains detailed information on experimental conditions and sources, facilitating the use of chemoinformatics and artificial intelligence. The findings were published in the Journal of Medicinal Chemistry.
Researchers at the Kurnakov Institute of General and Inorganic Chemistry of the Russian Academy of Sciences and Lomonosov Moscow State University, with financial support from Russia’s Ministry of Education and Science, developed the database. It contains more than 26,500 IC₅₀ values (a quantitative measure of cytotoxicity) for 7,050 complexes of five transition metals – ruthenium, iridium, rhodium, rhenium and osmium – tested on 754 cell lines. The data were manually collected from more than 1,900 scientific publications.
Some metals in complexes can penetrate cancer cells and destroy them – blocking division, damaging DNA or triggering cell death. A classic example is cisplatin, widely used in clinical practice. However, the search for new metal-containing drugs has long been hampered by a lack of systematic data. Existing international databases did not contain sufficient information on the action of such compounds, making it impossible to use artificial intelligence.
Using MetalCytoToxDB, the scientists trained machine learning models, GxP News reports. For ruthenium complexes, prediction accuracy (ROC‑AUC) reached 0.81; for iridium complexes, 0.73. An algorithm trained on studies published before 2024 was successfully tested on 2025 publications: in nine out of ten cases, the model correctly identified active compounds – twice the rate of random selection. A multi‑metal model was also developed for compounds with limited data.
The researchers noted that the current version of the models has limitations: they do not account for complex geometry (the arrangement of ligands around the metal atom), the influence of counterions, or selectivity toward healthy cells. These areas will be the subject of further research. Nevertheless, implementing the predictive model already makes it possible to narrow down candidate substances and reduce the pharmaceutical industry’s need for lengthy laboratory testing of ineffective molecules.
Earlier, it was reported that scientists at St. Petersburg State University created nanoparticles for slow and controlled drug delivery into the body. The development is expected to reduce the likelihood of side effects and cut the number of daily doses required for chronic diseases.