Capturing Unanticipated Drug Toxicities Using an Ensemble Machine Learning Approach.

Area of Science:

• Pharmacology
• Computational Chemistry
• Toxicology

Background:

• Many drugs are withdrawn post-market due to unforeseen toxicities despite pre-clinical safety assessments.
• Identifying drugs likely to cause adverse effects early in development is crucial for patient safety and reducing healthcare costs.

Purpose of the Study:

• To develop and validate a machine learning model for predicting drug withdrawal risk based on intrinsic drug properties.
• To identify key molecular and chemical features associated with unanticipated drug toxicities.

Main Methods:

• An ensemble machine learning classifier was trained using features including protein targets, protein structure, chemical fingerprints, and chemical properties.
• The model was evaluated using 10-fold cross-validation, achieving high accuracy and Matthews Correlation Coefficient.
• Feature importance analysis was conducted to identify key predictors of toxicity.

Main Results:

• The best-performing model achieved 92% accuracy and a 0.845 Matthews Correlation Coefficient in predicting drug withdrawal.
• Key predictive features included inhibition of cytochrome P450 enzymes and bile salt export pumps.
• The analysis highlighted both known and novel factors contributing to drug-induced toxicity.

Conclusions:

• Machine learning models can effectively predict drug withdrawal risk using pre-clinical data, reducing reliance on human trials for initial safety screening.
• Identifying novel toxicity predictors like bile salt export pump inhibition can guide safer drug design.
• This computational approach offers a promising strategy for enhancing drug safety evaluations during pharmaceutical development.