Artificial Intelligence and Machine Learning Models for Predicting Drug-Induced Kidney Injury in Small Molecules.

Area of Science:

• Pharmacology
• Computational Chemistry
• Drug Development

Background:

• Drug-induced kidney injury (DIKI) is a major obstacle in drug development, causing late-stage failures.
• Current predictive models often neglect crucial drug-target interactions, focusing solely on physicochemical properties.
• Early DIKI risk assessment is vital for enhancing drug safety and streamlining development.

Purpose of the Study:

• To develop an advanced AI/ML model for predicting DIKI risk.
• To integrate both physicochemical properties and off-target drug interactions for improved prediction accuracy.
• To create a tool for early screening of compounds with reduced DIKI potential.

Main Methods:

• A dataset of 360 FDA-classified compounds (129 nephrotoxic, 231 non-nephrotoxic) was curated.
• Physicochemical properties (55) and validated in vitro off-target interactions (6064) were analyzed.
• An ensemble machine learning model combining Ridge Logistic Regression, Support Vector Machine, Random Forest, and Neural Network was constructed.

Main Results:

• The ensemble model achieved an ROC-AUC of 0.86, with 0.79 sensitivity and 0.78 specificity.
• Key predictors included specific off-target interactions and physicochemical properties like PSA, pKa, and fsp3.
• The integrated approach effectively differentiated DIKI-inducing compounds from non-DIKI compounds.

Conclusions:

• Integrating physicochemical properties with off-target interaction data significantly enhances DIKI prediction accuracy.
• The developed AI/ML model serves as a valuable tool for early-stage identification of compounds with lower DIKI risk.
• This approach promises to improve drug safety and accelerate the drug development process.