An Algorithm Framework for Drug-Induced Liver Injury Prediction Based on Genetic Algorithm and Ensemble Learning.

Area of Science:

• Computational chemistry
• Toxicology
• Machine learning

Background:

• Drug-induced liver injury (DILI) is a significant challenge in drug discovery, leading to high attrition rates.
• Current computational methods for DILI prediction often struggle with high-dimensional and unbalanced datasets.
• Existing approaches using single molecular representations are insufficient, necessitating multi-modal fusion strategies.

Purpose of the Study:

• To develop an advanced computational framework for predicting DILI.
• To address the challenges of high dimensionality and data imbalance in DILI prediction.
• To enhance the accuracy and reliability of toxicity evaluation in drug discovery.

Main Methods:

• Integration of existing DILI datasets.
• Development of the Rotation-Ensemble-GA (R-E-GA) algorithm framework.
• Feature subset selection via rotation of fused molecular representation vectors.
• Integration of Adaboost-type ensemble learning to improve predictive performance.

Main Results:

• The R-E-GA algorithm demonstrated superior performance compared to state-of-the-art methods, including ensemble learning and graph neural network approaches.
• Five-fold cross-validation yielded promising results: an accuracy (ACC) of 0.77, an F1 score of 0.769, and an area under the curve (AUC) of 0.842.
• The proposed method effectively handles high-dimensional and unbalanced DILI prediction data.

Conclusions:

• The R-E-GA framework offers a robust and accurate approach for predicting drug-induced liver injury.
• This method enhances DILI prediction by effectively combining molecular representation fusion and ensemble learning.
• R-E-GA represents a significant advancement in computational toxicology for safer drug development.