Optimizing machine learning models for cytotoxicity prediction in lanthanide-doped nanomaterials: A data-driven approach for minimizing environmental hazards

Lanthanides (Lns) are vital to modern technology, with applications in advanced technologies such as computers, smartphones, and renewable energy systems. However, their potential side effects on biological systems remain unclear. Motivated by this concern, this study evaluated the impact of (Gd, Yb)-doped ZnO using in vitro cell models. The results indicate that the materials exhibited cytotoxicity and a tendency to agglomerate, which can lead to bioaccumulation within ecosystems, affecting both aquatic life and human health. To investigate the role of Ln in the bioactivity of ZnO, a machine learning (ML) analysis was conducted. A total of 23 key features of the materials were obtained via material characterization. Various regression models were trained and optimized, with the best model achieving an R² value exceeding 0.9 on the test set for predicting cell viability across the studied materials and cell cultures. Feature importance analysis was then performed to identify the most influential material features affecting biochemical activity. Concentration and cell type emerged as the most critical features contributing to improved model performance. These findings highlight the potential of ML as a robust computational tool for analyzing the cytotoxic properties of both doped and undoped inorganic nanomaterials. © 2025 Elsevier B.V.

Optimizing machine learning models for cytotoxicity prediction in lanthanide-doped nanomaterials: A data-driven approach for minimizing environmental hazards Academic Article in Scopus