XRepDDA: An Interpretable Drug-Disease Association Prediction Framework Leveraging Pretrained Chemical Language...

Area of Science:

• Computational drug discovery
• Pharmacology
• Bioinformatics

Background:

• Drug repositioning accelerates therapeutic development by finding new uses for existing drugs.
• Accurate prediction of drug-disease associations (DDAs) is crucial but challenged by poor drug representation, limited disease semantics, and imbalanced data.
• Existing computational methods struggle with predictive accuracy and generalization due to these limitations.

Purpose of the Study:

• To develop an innovative framework, XRepDDA, for enhanced DDA prediction.
• To improve the accuracy and robustness of computational drug repositioning strategies.
• To address limitations in drug representation, disease semantics, and data imbalance in DDA prediction.

Main Methods:

• Integrated multimodal feature representation using SMI-TED for drug embeddings and a hierarchical semantic graph (MeSH ontology) for disease representation.
• Employed deep metric learning with an improved ModernNCA architecture for discriminative embedding space learning.
• Utilized AllKNN adaptive undersampling to mitigate class imbalance and a multilevel explainability framework (SHAP, attention, molecular perturbation) for interpretability.

Main Results:

• XRepDDA significantly outperformed baseline models on benchmark datasets, achieving AUC and AUPR values up to 0.9990 and 0.9991.
• In silico validation via molecular docking supported the predictive reliability for Alzheimer's disease and stomach neoplasms.
• The explainability framework demonstrated chemical interpretability and biological plausibility of the predictions.

Conclusions:

• XRepDDA offers a robust and accurate computational framework for drug-disease association prediction.
• The integration of advanced representations and deep metric learning enhances drug repositioning efficiency.
• The developed explainability methods provide crucial insights into prediction mechanisms, supporting clinical translation.