icardioToxCSM

icardioToxCSM predicts six types of cardiotoxicity in small molecules to assess cardiac safety during drug discovery.

Key Features:

• Prediction endpoints: Predicts arrhythmia, cardiac failure, heart block, hERG toxicity, hypertension, and myocardial infarction for small molecules.
• Computational descriptors: Integrates graph-based signatures, molecular descriptors, toxicophore matchings, and molecular fingerprints.
• Machine learning: Combines the above features using explainable machine learning techniques to support interpretable predictions.
• Validation and performance: Validated with internal cross-validation schemes and external low-redundancy blind sets, achieving areas under ROC curves up to 0.898 in 5-fold cross-validation.
• Interpretability: Identifies substructures commonly associated with cardiotoxic compounds to provide mechanistic insight.

Scientific Applications:

• Early drug screening: Assess cardiotoxicity risks of new small-molecule candidates to inform compound prioritization and design of safer drugs.
• Retrospective analysis: Characterize cardiotoxic signatures in compounds withdrawn for cardiac issues, including examples such as fenspiride, propoxyphene, and valdecoxib.

Methodology:

Uses graph-based signatures, molecular descriptors, toxicophore matchings, and molecular fingerprints combined with explainable machine learning, and is evaluated via internal cross-validation (5-fold CV, AUC up to 0.898) and external low-redundancy blind sets.