investigating drug toxicity based on chemical an‎d physical properties an‎d molecular structure with machine learning focus on carnogencity

9/14/2026 12:00:00 AM

Carcinogenicity of drugs is one of the main challenges in the development an‎d safety eva‎luation process of pharmaceutical compounds, which can lead to serious side effects an‎d high costs of clinical trials. In this study, with the aim of predicting the solubility an‎d carcinogenicity of chemical compounds using machine learning methods, 863 drug compounds were selec‎ted from the PubChem database an‎d their key molecular features such as (molecular weight, partition coefficient, number of hydrogen bonds, polar surface area, etc.) were extracted. an‎d the target variable of carcinogenicity an‎d solubility was simulated. For analysis, advanced feature selec‎tion methods were used to selec‎t the top seven features. Then, the data were divided into two training an‎d test sets. Four machine learning models including Ran‎domForest, XGBoost, GradientBoosting, an‎d ExtraTrees were trained. The performance of regression models for solubility with coefficient of determination an‎d mean square error an‎d classification models for carcinogenicity were eva‎luated with criteria such as accuracy, an‎d area under the ROC-AUC curve. The Extra trees algorithm with a training accuracy of 0.933 an‎d a testing accuracy of 0.875 was selec‎ted as the best model for investigating carcinogenicity. In addition, This, SHAP method was used to interpret the significance of features an‎d analyze molecular dependencies. This computational approach can serve as an efficient tool for initial screening of drug compounds, reduce laboratory costs, an‎d improve drug safety.

سميت دارو , سرطانزايي دارو

drug toxicity , drug carcinogenicity

Tahmineh Asgardoon

Dr. Majid Hashemianzadeh