Predicting Cytotoxicity\nof Nanoparticles: A Meta-Analysis\nUsing Machine Learning

Cytotoxicity evaluation of nanoparticles\n(NPs) is regarded as a\ncrucial step for their successful application in the biomedical industry.\nHowever, conventional experimental methodologies for cytotoxicity\nmeasurements are often expensive, time-consuming, and demand intense\ntraining in cell culture. In this study, we developed generalized\nmachine learning (ML) models for both qualitative and quantitative\nprediction of cytotoxicity across a wide variety of NPs. In particular,\na meta-analysis of cytotoxicity data was conducted from published\nliterature on metallic, metal oxide, polymer, and carbon-based NPs,\nleading to the development of random forest-based regression and classification\nmodels for predicting cell viability from physicochemical properties\nof NPs, cellular attributes, and testing conditions. Our feature importance\nanalysis showed that accurately predicting the cytotoxicity of NPs\nusing the regression model requires knowledge of their composition,\nconcentration, zeta potential, and size, as well as exposure time,\ntoxicity assay, and tissue type. Interestingly, among these attributes,\nthe information about composition of NPs or tissue type was not needed\nfor achieving high accuracy in the qualitative prediction of cytotoxicity\nusing the classification model, indicating its superior robustness\ncompared to the regression model. These findings may encourage future\nresearchers to employ ML models more effectively and frequently to\nreliably assess the safety of NPs.