In silico Prediction of Drug Induced Liver Toxicity Using Substructure Pattern Recognition Method

Area of Science:

• Computational chemistry and toxicology
• Machine learning in drug discovery

Background:

• Drug-induced liver injury (DILI) is a significant cause of acute liver failure and drug market withdrawal.
• Early prediction of DILI is essential during drug discovery to mitigate risks.

Purpose of the Study:

• To develop high-accuracy classification models for predicting drug-induced liver toxicity.
• To identify structural alerts associated with DILI.

Main Methods:

• Collected a comprehensive dataset of 1317 diverse compounds.
• Employed five machine learning methods (SVM, etc.) using MACCS and FP4 fingerprints.
• Evaluated models using substructure pattern recognition and external validation.

Main Results:

• The best model, SVM with FP4 fingerprint, achieved 79.7% accuracy on training and 64.5% on test sets.
• External validation on 88 compounds from the Liver Toxicity Knowledge Base yielded 75.0% overall accuracy.
• Identified key substructure patterns correlated with drug-induced liver toxicity.

Conclusions:

• The developed SVM model demonstrates significant potential for predicting drug-induced liver toxicity.
• The identified structural alerts can aid in designing safer drug candidates.