toxCSM

toxCSM predicts toxicity profiles of small molecules to support early-stage screening and optimization in drug and agrochemical development.

Key Features:

• Comprehensive Toxicity Prediction: Predicts a wide range of toxicity properties for small molecules relevant to pharmaceutical and agrochemical safety.
• Graph-Based Signatures and Molecular Descriptors: Employs graph-based signatures and molecular descriptors to represent chemical structures and capture features related to toxicity.
• Similarity Scores: Uses similarity scores to assess likeness of new compounds to known toxic substances for adverse-effect prediction.
• Model Development: Implements 36 distinct predictive models targeting various toxicity endpoints.
• Performance Metrics: Reports evaluation metrics including Area Under the Receiver Operating Characteristic Curve (AUC) up to 0.99 and Pearson's correlation coefficients up to 0.94 based on 10-fold cross-validation and blind test sets.

Scientific Applications:

• Early-Stage Toxicity Screening: Enables prioritization of chemical compounds by predicted toxicity during lead identification and optimization.
• Drug Discovery: Supports the identification and optimization of safer drug candidates by predicting human-relevant toxicity endpoints.
• Agrochemical Development: Assists in evaluating environmental and human toxicity risks of agrochemical candidates.
• Toxicity Profile Optimization: Facilitates comparison of analogues to guide chemical modifications that reduce predicted toxicity.

Methodology:

Models were developed using graph-based signatures, molecular descriptors and similarity scores, with 36 models evaluated by 10-fold cross-validation and tested on blind test sets reporting AUC and Pearson metrics.