Background and aims: Nanoparticles have attracted considerable attention in biomedical applicationsdue to their unique characteristics. However, the
toxicity of nanoparticles remains aconcern, especially for use in biological systems. The biological
toxicity of nanoparticles can leadto various responses such as apoptosis, inflammation, allergies, neurotoxicity, fibrosis, hematologicaltoxicity, pulmonary toxicity, carcinogenicity, and genotoxicity. Since the experimentalanalysis of
toxicity is time-consuming and costly, an alternative approach, such as artificial intelligencetechniques, could be useful to predict the
toxicity of nanoparticles.Methods: A dataset containing ۲۴۶ records of
nanoparticle properties, along with the effect onvarious cell lines, was retrieved from the NanoHub repository in ۲۰۲۲. After preprocessing thedataset using K-Nearest Neighbors for imputing missing values, Gini Index was used to analyzethe effects of different factors on
nanoparticle toxicity. The ۱۰-Fold cross-validation was used tobuild and evaluate a Random Forest model for the classification of nanoparticles. Hyperparametersof the Random Forest model were optimized by a grid search of parameters using the traindata in each fold. The model’s performance was evaluated in terms of accuracy, sensitivity, specificity,F-measure, and area under the receiver operating characteristic curve.Results: Based on Gini Index, the cell type was the most correlated factor with toxicity, followedby exposure dose, tissue,
nanoparticle type, specific surface area, and surface coatings. TheRandom Forest model predicted the
toxicity of nanoparticles with a ۹۳.۴۵% (±۳.۳۹%) accuracy,۹۲.۷۰% (±۸.۲۰%) sensitivity, and ۹۴.۱۸% (±۵.۶۷%) specificity. Also, the F-measure of the RandomForest Model was equal to ۹۲.۴۴% (±۳.۸۵%), and the AUC of the ROC was equal to ۰.۹۶۶(±۰.۰۲۷).Conclusion: Compared to similar studies, the findings obtained in this study have provided satisfactoryresults, indicating this momodel’sigh performance. Artificial intelligence techniques couldbe useful in the prediction of
nanoparticle toxicity that results in omitting excessive laboratorywork. However, more data is required to create robust models for predicting
nanoparticle toxicity.