AI/ML Models to Predict the Severity of Drug-Induced Liver Injury for Small Molecules

Area of Science:

• Pharmacology and Toxicology
• Computational Chemistry
• Artificial Intelligence in Drug Discovery

Background:

• Drug-induced liver injury (DILI) is a major challenge in drug development, leading to costly attrition.
• Existing predictive models often overlook liver-expressed proteins and drug interactions.
• Early identification of DILI risk is essential to reduce development costs and timelines.

Purpose of the Study:

• To develop an integrated artificial intelligence/machine learning (AI/ML) model for predicting DILI severity in small molecules.
• To improve DILI risk assessment by incorporating physicochemical properties and in silico predicted off-target interactions.
• To address the gap in current predictive models by including liver-expressed protein targets.

Main Methods:

• Compiled a dataset of 603 compounds categorized by DILI severity (Most DILI, Less DILI, No DILI) from FDA data.
• Developed a consensus AI/ML model using six machine learning methods (SVM, RF, LR, WA, PLR, k-NN, NB, ANN).
• Integrated physicochemical properties (e.g., fsp3, log S) and predicted off-target interactions (e.g., PTGS1, CYP2C9) for prediction.

Main Results:

• The consensus model achieved a receiver operating characteristic area under the curve of 0.88, with 0.73 sensitivity and 0.9 specificity in identifying Most DILI and No DILI compounds.
• Physicochemical properties and approximately 43 identified off-targets were significant predictors of DILI severity.
• Key identified off-targets include PTGS1, PTGS2, PPARγ, RXRA, and CYP2C9.

Conclusions:

• The integrated AI/ML approach significantly enhances DILI predictivity compared to models using only chemical properties.
• Combining physicochemical data with predicted biological interactions offers a more robust method for DILI risk assessment.
• This computational strategy can aid in early-stage drug discovery and development by identifying potential DILI liabilities.