Development of method using language processing techniques for extracting information on drug-health food product...

Area of Science:

• Pharmacology and Bioinformatics
• Drug-Food Interactions
• Computational Biology

Background:

• Health food products (HFPs) can cause adverse drug interactions.
• Information on drug-HFP interactions (DHIs) is crucial for patient safety.
• Existing resources for DHIs are often fragmented.

Purpose of the Study:

• To establish an efficient workflow for extracting Drug-HFP Interactions (DHIs) from open resources.
• To integrate network analysis and biomedical context for DHI prediction.
• To identify both known and potentially unknown DHIs.

Main Methods:

• Collected data on drugs, enzymes, interactions, and known DHIs from public databases and literature.
• Constructed a network of enzymes, HFPs, and drugs, identifying enzymes as interaction hubs.
• Applied BioWordVec to analyze biomedical context and predict potential DHIs.

Main Results:

• Identified 44,965 DHIs (30% known) in the initial network analysis.
• Method 2-1 yielded 7401 DHIs (17% known) by focusing on enzyme-specific HFPs.
• Method 2-2 identified 2819 DHIs (30% known) by selecting HFPs with similar functions to drugs.

Conclusions:

• An integrated workflow combining language processing and network analysis was successfully constructed.
• This workflow enables efficient extraction of a wide range of known and unknown DHIs.
• The developed method enhances the identification of potential drug-health food product interactions.