Machine learning-based optimization of cytotoxicity testing for assessing Zn-based biodegradable metals

Zinc (Zn)-based biodegradable metals are emerging as promising candidates for biomedical implants. Nonetheless, discrepancies between in vitro and in vivo biocompatibility findings for these metals often complicate their evaluation. This study aims to optimize cytotoxicity testing for Zn-based metals using machine learning techniques. Data from 51 cytotoxicity experiments on pure Zn were utilized to train and refine five predictive models, i.e., decision tree (DT), random forest, gradient boosted decision tree, support vector machine, and multilayer perceptron (MLP). In addition, the impact of pure Zn samples on the viability of bone-related cells, endothelial cells, and fibroblasts was assessed. The models were optimized for comparable performance, with the MLP model indicating that at concentrations below 40 %, all cell types demonstrate a high probability of non-toxicity. The "Extract concentration" by the DT model was a critical predictive factor. Cytotoxicity tests confirmed that the cell survival rates remained high at Zn extract concentrations up to 40 %, beyond which cell viability significantly declined. This research offers innovative insights into the cytotoxicity testing protocols for Zn-based biomaterials, elucidating key factors that affect cytotoxicity assessments and defining the limits of in vitro evaluations. Lastly, this study enhances the reliability of toxicity assessments and supports the development of a standardized framework for evaluation metrics.