Ontology-Enhanced Deep Learning for Mechanistic Prediction of Drug-Drug Interactions: A Clinically Interpretable...

Area of Science:

• Pharmacology
• Bioinformatics
• Computational Biology

Background:

• Drug-drug interactions (DDIs) significantly impact patient safety and healthcare.
• Predicting DDIs and their biological mechanisms is crucial but challenging.
• Existing methods often lack mechanistic interpretability.

Purpose of the Study:

• To enhance the accuracy and mechanistic interpretability of DDI prediction.
• To leverage biomedical ontologies for biologically enriched drug feature representation.
• To integrate ontology-based embeddings with deep learning for improved DDI prediction.

Main Methods:

• Developed ontology-based embeddings using SIDER, DrugBank, and Gene Ontology data.
• Represented phenotypic, functional, and mechanistic drug characteristics.
• Employed neural networks for DDI prediction using these enriched features.

Main Results:

• Achieved robust DDI prediction performance, with external validation yielding AUC values up to 0.94.
• Successfully predicted DDIs across 11 pharmacokinetic and pharmacodynamic mechanisms.
• Identified high-risk DDI mechanisms linked to known adverse drug reactions, enhancing interpretability.

Conclusions:

• The integration of ontology-based embeddings with deep learning significantly improves DDI prediction accuracy.
• The approach provides mechanistic insights, supporting clinical decisions to mitigate adverse drug reaction risks.
• This method offers a biologically and pharmacologically relevant framework for understanding and predicting DDIs.