Mixed learning algorithms and features ensemble in hepatotoxicity prediction.

Area of Science:

• Pharmacology
• Toxicology
• Computational Chemistry

Background:

• Drug-induced liver injury (DILI) is a significant safety concern in drug development and patient treatment.
• Predicting DILI is challenging due to its complex mechanisms and infrequent occurrence.

Purpose of the Study:

• To develop and validate a robust computational model for predicting drug-induced liver injury (hepatic effects).
• To create an ensemble machine learning model that leverages diverse algorithms and features for improved prediction accuracy.

Main Methods:

• An ensemble model comprising 617 base classifiers was constructed using a dataset of 1,087 diverse compounds.
• Model validation involved five-fold cross-validation and 25 rounds of y-randomization for internal assessment.
• External validation was performed on an independent set of 120 compounds.

Main Results:

• The ensemble model achieved 75.0% accuracy, 81.9% sensitivity, and 64.6% specificity in external validation.
• The model successfully identified 22 out of 23 withdrawn drugs or those with black box warnings for hepatotoxicity.
• Dronedarone, a drug associated with severe liver injury, was correctly predicted as hepatotoxic.

Conclusions:

• The developed ensemble model demonstrates strong predictive capability for identifying hepatotoxic compounds.
• The model shows potential for enhancing drug safety assessments and reducing drug development failures.
• The ensemble model is publicly accessible for use in predicting drug-induced liver injury.