ToxinPredictor: Computational models to predict the toxicity of molecules.

Area of Science:

• Computational Chemistry
• Toxicology
• Machine Learning

Background:

• Predicting molecular toxicity is crucial for drug discovery, environmental protection, and chemical management.
• Traditional experimental toxicity testing is resource-intensive and time-consuming.
• Computational models offer a faster, more cost-effective alternative for toxicity assessment.

Purpose of the Study:

• To develop and validate ToxinPredictor, a machine learning model for predicting small molecule toxicity.
• To identify key molecular descriptors influencing toxicity predictions.
• To provide a publicly accessible webserver for toxicity prediction.

Main Methods:

• A Support Vector Machine (SVM) model was developed using curated datasets of toxic and non-toxic molecules.
• Feature selection techniques, including Boruta and Principal Component Analysis (PCA), were employed.
• SHapley Additive exPlanations (SHAP) analysis was used for model interpretability.

Main Results:

• The SVM-based ToxinPredictor achieved high performance with an Area Under the Receiver Operating Characteristic curve (AUROC) of 91.7%, an F1-score of 84.9%, and an accuracy of 85.4%.
• The model outperformed existing computational toxicity prediction solutions.
• SHAP analysis identified critical molecular descriptors contributing to toxicity predictions.

Conclusions:

• ToxinPredictor offers a reliable and accurate computational framework for assessing molecular toxicity.
• The model enhances safety in drug development and environmental health evaluations.
• A user-friendly webserver is available to facilitate the prediction of toxic compounds.