Machine Learning for Predicting Human Drug-Induced Cardiotoxicity: A Scoping Review

Area of Science:

• Pharmacology and Toxicology
• Computational Biology
• Drug Development

Background:

• Drug-induced cardiotoxicity is a significant hurdle in pharmaceutical research and patient safety.
• Existing machine learning (ML) approaches often focus narrowly on specific mechanisms like hERG inhibition.
• A broader predictive scope is needed for comprehensive cardiotoxicity assessment.

Purpose of the Study:

• To systematically review studies utilizing ML models for predicting a wide spectrum of drug-induced cardiotoxicity.
• To identify common data sources, features, algorithms, and performance metrics in this field.
• To assess the current state and potential of ML in drug cardiotoxicity prediction.

Main Methods:

• Systematic literature search across major scientific databases (PubMed, EMBASE, SCOPUS, Web of Science).
• Extraction and categorization of data including study sources, feature types, ML algorithms, and performance evaluation methods.
• Analysis of 25 selected studies meeting inclusion criteria for ML-based cardiotoxicity prediction.

Main Results:

• Studies covered diverse cardiotoxicity outcomes like arrhythmia, cardiac failure, and myocardial infarction.
• SIDER database and molecular descriptors were common data sources and features.
• Support Vector Machines (SVM) and Random Forest (RF) were frequently used, demonstrating promising predictive performance with AUC-ROC > 0.70 and accuracy > 0.75 in several cases.
• External validation was limited but showed ML's potential.

Conclusions:

• Machine learning holds substantial promise for predicting various drug-induced cardiotoxicity.
• Integrating diverse data types and employing robust validation strategies are key to enhancing predictive models.
• Further research is needed to address limitations in external validation and improve model generalizability.