Enhancing drug-food interaction prediction with precision representations through multilevel self-supervised...

Area of Science:

• Pharmacology and Computational Chemistry
• Artificial Intelligence in Healthcare

Background:

• Drug-food interactions (DFIs) significantly affect drug efficacy and patient safety by altering pharmacokinetics.
• Current computational models for predicting DFIs are limited by the complexity and incomplete data of food compounds.

Purpose of the Study:

• To develop an advanced computational model, DFI-MS, for accurate prediction of drug-food interactions.
• To overcome challenges in food component data acquisition and feature representation in DFI research.

Main Methods:

• DFI-MS utilizes novel modules for perturbation interactions, feature alignment, domain separation, and inference feedback.
• Employs data augmentation, contrastive learning, and a feature interaction encoder for robust feature representation.
• Addresses diverse data frequencies and characteristics through feature alignment and domain separation.

Main Results:

• DFI-MS achieves precise characterization of food features, a significant advancement in DFI research.
• Demonstrates exceptional performance across multiple benchmark datasets.
• The model offers flexibility for various downstream tasks via its inference feedback module.

Conclusions:

• DFI-MS represents a substantial progress in computational DFI prediction by improving food feature representation.
• The developed methodology enhances the accuracy and reliability of predicting drug-food interactions, benefiting patient safety.