Beyond Molecular Structures: Investigating Demographic Factors in Drug-Induced Cardiotoxicity Prediction Models.

Area of Science:

• Pharmacology
• Toxicology
• Computational Biology

Background:

• Drug-induced cardiotoxicity is a major safety concern, causing trial failures and drug withdrawals.
• Existing models struggle to incorporate demographic factors (sex, age, body mass) for personalized risk prediction.

Purpose of the Study:

• To develop and evaluate the CARBIDE (CARdiotoxicity Based on Integrated Demographic Evidence) framework.
• To assess if structure-demographic interactions can be learned from spontaneous pharmacovigilance data for cardiotoxicity prediction.

Main Methods:

• Created 27 dataset variants from the FAERS database.
• Systematically evaluated filtering criteria, cardiotoxicity definitions, and statistical methods.
• Conducted ablation studies to assess the impact of demographic features.

Main Results:

• Machine learning models failed to learn meaningful structure-demographic relationships.
• Demographic features provided little additional predictive value beyond structural information.
• Models primarily learned population-level statistics or relied solely on chemical structure.

Conclusions:

• Current pharmacovigilance data has fundamental limitations for demographic-aware toxicity prediction.
• CARBIDE provides insights into these limitations, guiding future personalized cardiotoxicity prediction efforts.