Computational Models for Human and Animal Hepatotoxicity with a Global Application Scope

Area of Science:

• Pharmacology and Toxicology
• Computational Chemistry
• Drug Discovery

Background:

• Drug-induced liver injury (DILI) is a major cause of drug attrition, difficult to predict with current preclinical models.
• Existing computational models often fail to cover the chemical space of novel drug candidates.
• Improved in-silico methods are urgently needed for early identification of safe drug candidates.

Purpose of the Study:

• To develop and validate advanced in-silico models for predicting drug-induced liver toxicity.
• To enhance the reliability of computational approaches in drug discovery and development.
• To integrate predictive models into early safety workflows for experimental design support.

Main Methods:

• Collected a large dataset of 3712 compounds with human and animal liver toxicity findings.
• Clustered hepatotoxicity into 21 detailed preclinical and human endpoints.
• Trained support vector machine models using optimized descriptors and validated with internal and external datasets.

Main Results:

• The global human hepatotoxicity model demonstrated high sensitivity (68%) and excellent specificity (95%) on an internal validation set.
• Models for preclinical endpoints showed similar performance.
• Predictions were tagged with confidence parameters derived from statistical analysis for external validation.

Conclusions:

• The developed in-silico approach provides a reliable method for predicting drug-induced liver toxicity in novel compounds.
• The models are integrated into an early safety workflow, aiding in the identification of safer drug candidates.
• This approach addresses the limitations of current computational models by improving chemical space coverage and prediction accuracy.