Computational models for predicting liver toxicity in the deep learning era.

Area of Science:

• Pharmacology and Toxicology
• Computational Chemistry
• Artificial Intelligence in Medicine

Background:

• Drug-induced liver injury (DILI) is a critical safety concern, potentially leading to severe outcomes including liver failure and death.
• Quantitative structure-activity relationship (QSAR) models are vital for early hepatotoxicity screening due to their non-physical substance requirements and speed.
• Recent advancements in deep learning (DL) have enabled the development of sophisticated QSAR models.

Purpose of the Study:

• To review the application of deep learning (DL) in predicting drug-induced liver injury (DILI).
• To focus on the development of QSAR models utilizing extensive chemical structure datasets and DILI outcomes.
• To evaluate DL methods against traditional machine learning (ML) approaches for DILI prediction.

Main Methods:

• Comprehensive review of deep learning (DL) methodologies applied to DILI prediction.
• Analysis of QSAR models developed using chemical structure data and DILI outcomes.
• Comparative evaluation of DL techniques versus traditional machine learning (ML) approaches.

Main Results:

• Deep learning (DL) models show significant potential for enhancing the accuracy and efficiency of DILI prediction.
• Comparison highlights the strengths and limitations of DL techniques in terms of interpretability, scalability, and generalization.
• DL-based QSAR models offer a promising avenue for early-stage hepatotoxicity screening.

Conclusions:

• Deep learning methodologies are poised to significantly improve DILI risk prediction.
• Future research should focus on leveraging DL for robust predictive models to mitigate DILI in humans.
• Enhanced predictive models can contribute to safer drug development and improved patient outcomes.