toxCSM: comprehensive prediction of small molecule toxicity profiles

Area of Science:

• Computational toxicology
• Drug discovery and development
• Environmental and human health risk assessment

Background:

• Drug discovery is expensive and prone to failure, with toxicity being a primary reason for clinical trial attrition.
• Existing computational tools for toxicity prediction are limited in scope and scale.
• There is a need for comprehensive platforms to assess human and environmental toxicity early in drug development.

Purpose of the Study:

• To present toxCSM, a computational platform for studying and optimizing the toxicity profiles of small molecules.
• To provide a tool for early identification and optimization of toxicity, thereby reducing drug development costs and timelines.
• To develop predictive models for a wide range of toxicity properties relevant to drug and agrochemical safety.

Main Methods:

• Leveraging graph-based signatures, molecular descriptors, and similarity scores.
• Developing 36 predictive models for various toxicity endpoints.
• Utilizing 10-fold cross-validation and blind test sets for performance evaluation.

Main Results:

• Achieved high predictive performance with Area Under the Receiver Operating Characteristic (ROC) Curve (AUC) up to 0.99 and Pearson's correlation coefficients up to 0.94.
• Demonstrated superior performance compared to alternative methods on both cross-validation and blind test sets.
• Developed a comprehensive suite of models for diverse toxicity assessments.

Conclusions:

• toxCSM offers a robust computational solution for predicting small molecule toxicity.
• The platform can significantly assist in developing safer drugs and agrochemicals by enabling early toxicity profiling.
• toxCSM is accessible via a user-friendly web server and API for broad scientific use.