NanoTox: Development of a Parsimonious <i>In Silico</i> Model for Toxicity Assessment of Metal-Oxide Nanoparticles Using Physicochemical Features

Metal-oxide nanoparticles find widespread applications in mundane life today, and cost-effective evaluation of their cytotoxicity ecotoxicity is essential for sustainable progress. Machine learning models use existing experimental data learn quantitative feature–toxicity relationships to yield predictive models. In this work, we adopted a principled approach problem by formulating novel feature space based on intrinsic extrinsic physicochemical properties, including periodic table properties but exclusive vitro characteristics such as cell line, type, assay method. An optimal hypothesis was developed applying variance inflation analysis the correlation structure features. Consequent stratified train-test split, training dataset balanced toxic outcomes mapping then achieved from normalized toxicity class using various hyperparameter-tuned machine models, namely, logistic regression, random forest, support vector machines, neural networks. Evaluation an unseen test set yielded >96% accuracy network with one-hidden-layer The obtained are parsimonious, intelligible inputs, embedded applicability check. Interpretability investigations identified key predictor variables metal-oxide nanoparticle cytotoxicity. Our could be applied new, untested oxides, majority-voting ensemble classifier, NanoTox, that incorporates best above NanoTox first open-source nanotoxicology pipeline, freely available under GNU General Public License (https://github.com/NanoTox).