MolToxPred: small molecule toxicity prediction using machine learning approach.

Area of Science:

• Computational chemistry
• Toxicology
• Machine learning in drug discovery

Background:

• Drug development faces challenges due to chemical toxicity, increasing costs and time.
• Machine learning (ML) offers a solution for predicting toxicity, reducing experimental needs and ethical concerns.

Purpose of the Study:

• To develop MolToxPred, an ML-based tool for predicting the toxicity of small molecules and metabolites.
• To enhance efficiency and safety in drug discovery and chemical assessment.

Main Methods:

• Developed a stacked ML model using Random Forest, Multi-layer Perceptron, and LightGBM as base classifiers, with Logistic Regression as the meta-classifier.
• Utilized diverse molecular descriptors and fingerprints, coupled with feature selection and Bayesian optimization.
• Employed stratified 5-fold cross-validation for training and validation.

Main Results:

• MolToxPred achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 87.76% on the test set and 88.84% on an external validation set.
• The stacked model demonstrated superior performance compared to individual base classifiers and an existing tool.
• Identified structural alerts associated with toxicity endpoints for pathway analysis.

Conclusions:

• MolToxPred provides a robust and efficient in silico method for toxicity prediction.
• The tool can aid drug discovery and regulatory processes by minimizing experimental toxicity testing.
• The stacked ensemble approach and comprehensive feature engineering contribute to its high predictive accuracy.