A multi-label learning model for predicting drug-induced pathology in multi-organ based on toxicogenomics data

Area of Science:

• Toxicogenomics
• Computational Toxicology
• Drug Development

Background:

• Drug-induced toxicity is a major cause of drug withdrawal and necessitates early identification during development.
• Current methods often focus on single organs or binary toxicity, limiting comprehensive assessment.
• Detailed pathological findings are crucial for accurate toxicity evaluation.

Purpose of the Study:

• To develop a novel multi-label learning model, Att-RethinkNet, for predicting drug-induced pathological findings in the liver and kidney.
• To improve the accuracy and reliability of early-stage toxicity prediction using toxicogenomics data.
• To create a more comprehensive and generalized model applicable to multiple organs and factors like dose and administration time.

Main Methods:

• Proposed Att-RethinkNet, a multi-label learning model incorporating a memory structure and attention mechanism.
• Utilized toxicogenomics data for predicting pathological findings.
• The model considers compound type, dose, and administration time for enhanced generalization.
• Enabled simultaneous prediction of multiple pathological findings.

Main Results:

• Att-RethinkNet demonstrated competitive performance compared to state-of-the-art methods.
• The model accurately predicts potential hepatotoxicity and nephrotoxicity.
• Achieved more reliable predictions of multiple pathological findings concurrently.

Conclusions:

• Att-RethinkNet offers a significant advancement in predicting drug-induced hepatotoxicity and nephrotoxicity.
• The model's ability to predict multiple pathologies simultaneously enhances early drug safety assessment.
• This approach provides a more comprehensive and reliable tool for toxicogenomics-based drug development.