Introduction
The complexity of food-drug interactions (FDIs) presents a significant challenge for clinical practice. While warnings like "do not take with grapefruit juice" are well-known, the vast majority of FDI data remains trapped in unstructured scientific literature or semi-structured databases. Previous attempts to automate data extraction using Natural Language Processing (NLP) have yielded low accuracy (approximately 55% F-score), making them unreliable for clinical decision support.
To address this, researchers have significantly upgraded the Food Interactions with Drugs Evidence Ontology (FIDEO). By integrating over 1,700 structured interactions from expert-curated databases, the team has created a FAIR-compliant (Findable, Accessible, Interoperable, Reusable) resource that provides a reliable "knowledge map" for navigating dietary-medical relationships.
Methodology: The XOD Workflow
The enrichment of FIDEO followed a semi-automated workflow based on eXtensible Ontology Development (XOD) principles. This approach ensures that the ontology remains maintainable and interoperable with global standards.
The process was divided into four primary tasks:
- Term Annotation: Manual annotation of textual descriptions from DrugBank to identify specific foods and herbs.
- Ontology Term Reuse (XOD 1): Mapping terms to existing gold-standard ontologies such as ChEBI (chemicals), FoodOn (food products), and DrOn (drug products) via the Ontology Lookup Service (OLS).
- Semantic Alignment (XOD 2): Organizing drugs using the Anatomical Therapeutic Chemical (ATC) classification system to provide a clear hierarchy.
- Logical Definition Generation (XOD 3): Using the ROBOT tool to transform spreadsheet-based templates into complex Web Ontology Language (OWL) axioms. This "Ontology-as-a-Spreadsheet" approach simplifies updates and maintenance.
Data Integration and Results
The study integrated data from two primary expert-curated sources: DrugBank (v5.0) and Hedrine (a French resource for herb-drug interactions). By prioritizing curated data over raw text mining, the authors ensured a higher foundation of reliability.
The updated FIDEO ontology now includes:
- 1,702 unique food-drug interactions.
- 721 distinct drugs and 138 foods/herbs.
- 885 information sources linked directly to the evidence.
This structured data allows for complex "Competency Questions." For example, the ontology can identify that 29 different antithrombotic agents (including heparin and rivaroxaban) may interact with garlic. The data also highlights major interaction "hubs," such as alcoholic beverages (326 interactions) and warfarin (17 distinct food/herb interactions).
Structural Strategy: Precompiled Concepts
A key innovation in FIDEO is the use of "precompiled concepts." Rather than simply linking a food node to a drug node with a generic relationship, each interaction is represented as a unique, structured entity.
As shown in the figure, an interaction (e.g., garlic-warfarin) is defined as a process with specific inputs (the chemical entity and the food product) and is linked to an information source. This granular representation enables powerful automated reasoning and more precise querying than traditional flat data structures.
Limitations and Future Directions
While FIDEO represents a major step forward, several challenges remain:
- Standardization: Food descriptions in source databases are often inconsistent (e.g., "aged cheese" vs. "tyramine-containing foods").
- Mechanistic Data: Current versions do not always capture the biological mechanism behind an interaction.
- Evidence Grading: The ontology does not yet distinguish between interactions observed in humans versus those found in animal or in vitro studies.
Future work will focus on addressing these gaps and leveraging the structured data to train machine learning models capable of predicting unknown interactions, moving from a reactive record to a proactive clinical tool.
The new version of FIDEO is open-source and available on GitHub.
Summary updated