CORAL: Binary classifications (active/inactive) for drug-induced liver injury

Area of Science:

• Computational chemistry
• Toxicology
• Drug discovery

Background:

• Human hepatotoxicity, or drug-induced liver injury (DILI), is critical in drug discovery.
• Experimental data on DILI is limited, posing challenges for drug development.
• Quantitative structure-activity relationships (QSAR) offer a promising alternative to experimental methods.

Purpose of the Study:

• To develop predictive QSAR models for human hepatotoxicity.
• To utilize the Monte Carlo method and CORAL software for model development.
• To assess the predictability and stability of the developed DILI models.

Main Methods:

• Development of binary classification QSAR models using the Monte Carlo method and CORAL software.
• Employing simplified molecular input line entry systems (SMILES) and hydrogen suppressed graphs (HSG) for molecular representation.
• Validation using external datasets and stability assessment through random splits into training and validation sets.

Main Results:

• Models achieved Matthew's correlation coefficients ranging from 0.52 to 0.62 on external validation sets.
• Stochastic experiments confirmed the stability and predictability of the models across various data splits.
• The QSAR approach demonstrated good predictive ability for drug-induced liver injury.

Conclusions:

• QSAR modeling, particularly using Monte Carlo techniques with CORAL software, is effective for predicting drug-induced liver injury.
• This computational approach provides a valuable tool for extending DILI databases and aiding drug discovery.
• The developed models show stability and good predictive performance, supporting their utility in toxicological assessments.