=Paper=
{{Paper
|id=Vol-2807/paperL
|storemode=property
|title=FIDEO: Food Interactions with Drugs Evidence Ontology
|pdfUrl=https://ceur-ws.org/Vol-2807/paperL.pdf
|volume=Vol-2807
|authors=Georgeta Bordea,Jean Nöel Nikiema,Romain Griffier,Thierry Hamon,Fleur Mougin
|dblpUrl=https://dblp.org/rec/conf/icbo/BordeaNGHM20
}}
==FIDEO: Food Interactions with Drugs Evidence Ontology==
https://ceur-ws.org/Vol-2807/paperL.pdf
September 2020
FIDEO: Food Interactions with Drugs
Evidence Ontology
Georgeta BORDEA a,1 , Jean Nöel NIKIEMA b , Romain GRIFFIER a ,
Thierry HAMON c,d , and Fleur MOUGIN a
a Univ. Bordeaux, Inserm UMR 1219, Bordeaux Population Health, ERIAS, France
b Centre de recherche du CHUM, Montreal, Qc, Canada
c LIMSI, CNRS UPR 3251, Université Paris-Saclay, Orsay, France
d Université Paris 13, Sorbonne Paris Cité, Villetaneuse, France
Abstract. In this paper we propose the Food Interactions with Drugs Evidence On-
tology (FIDEO), an ontology used for annotation and retrieval of scientific articles
about food-drug interactions. Currently available ontologies address mainly drug-
drug interactions, but much less attention has been given to clinically significant
food-drug interactions. This work proposes an extension of a drug interaction on-
tology following the METHONTOLOGY methodology with the goal of represent-
ing potential drug interactions with foods, food components and food categories.
To evaluate the proposed formal ontological model, we discuss the results of pop-
ulating the ontology with information from manually annotated abstracts and from
a compendium.
Keywords. Biomedical Ontology, Food-Drug Interactions, Text Mining, Adverse
Drug Effects
1. INTRODUCTION
In clinical practice convenience, safety and reduced costs are some of the reasons that
make the oral route a preferred method for drug administration. But a main limitation
of this route of administration is related to the way drugs move through the digestive
tract because of their likely association with other drugs and foods, which may affect
how much and how fast a drug is absorbed. To prevent undesired interactions, clini-
cal trials are required prior to drug marketing to systematically analyse the absorption
of a drug with respect to standard meals. Additionally, chemical substances and com-
pounds contained in specific foods may occasionally interact with drugs, dramatically
increasing or reducing the effect of a drug. Grapefruit for example contains bioactive
furocoumarins and flavonoids that activate or deactivate many drugs in ways that can be
life-threatening [8]. With a rapid increase in the number of biomedical publications that
report food-drug interactions [4], there is a growing need for systems that automatically
process scientific literature and for formal models to represent this information. The Food
Interactions with Drugs Evidence Ontology, namely FIDEO2 , is created in the frame of
the French ANR MIAM (Maladies, Interactions Alimentation-Médicaments, 2017-2020)
1 Corresponding Author: Georgeta Bordea E-mail: name.surname@u-bordeaux.fr.
2 FIDEO Git: https://gitub.u-bordeaux.fr/erias/fideo.git
Copyright © 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
September 2020
project3 aiming to automatically extract interactions between foods and drugs and to rep-
resent them in an ontology. More precisely, the objectives of this research project are:
(i) to automatically identify mentions of food-drug interactions in scientific articles using
text analysis, (ii) to represent these interactions in a formal way, and (iii) to make them
accessible to healthcare professionals within the Thériaque R database4 , which contains
exhaustive information related to drugs marketed in France [19]. This widely used na-
tional database, curated by pharmacists, is independent from pharmaceutical companies
and the national healthcare insurance fund, gathering information from official sources
and from reference books. In this work, we address the second objective of the project
and we describe the resulting ontology and the methodology followed to design it.
This paper is organised as follows: we first give an overview of the related work in
section 2, than we describe the FIDEO requirements in section 3. We discuss the materi-
als used to engineer the ontology in section 4 and we present the followed methodology
in section 5. We conclude this work with a discussion in section 6 and conclusions in
section 7.
2. RELATED WORK
The FIDEO ontology builds on one side on large scale efforts modelling the foods do-
main, namely the FoodOn ontology [9], and on another side on recent work in build-
ing evidence-driven ontologies for potential drug-drug interactions, that is the DIDEO
ontology [7]. Both ontologies are based on the Basic Formal Ontology, BFO [12] and
are compliant with the OBO Foundry [25] principles5 . FoodOn has been adopted by
health researchers and is extended by the Ontology for Nutritional Epidemiology (ONE)
with nutritional epidemiology concepts, to assesses the relations between diet, nutrients
and health, and disease outcomes [26]. An annotated corpus of online recipes is avail-
able [23] that uses FoodIE, an approach for named entity recognition which identifies
food items in text [22]. Although FoodOn is a large and growing resource about foods,
the types of foods described are far from complete and lag behind information contained
in Wikipedia. Recent work proposes a dataset for extracting domain-specific information
about foods from a knowledge graph [3]. DIDEO takes an evidence-driven approach to
representing drug-drug interactions, acknowledging the need to provide users with sup-
porting evidence to allow them to assess the clinical importance of an interaction and to
make management decisions [6]. The type of study that describes an interaction is also
represented in the ontology, including for example in vitro experiments, population phar-
macokinetic analyses, randomised controlled clinical trials, and observational epidemio-
logic studies. In [7], DIDEO developers point out that the main drawbacks of the Drug In-
teraction Ontology (DIO) [27] are the following: (i) it contains inconsistencies, and (ii) it
does not specify how main entities differ (i.e., drugs, chemicals and molecules). Com-
pared to DIDEO, the Drug-Drug Interactions Ontology (DINTO) [17] received higher
traction from the ontology learning community mainly because of a shared task on drug-
drug interaction extraction [24] and the associated annotated corpus [18]. But a main
limitation of this ontology is that it is not designed from the start based on the BFO on-
tology, although a concept alignment to BFO is proposed. DINTO does not allow us to
3 MIAM project: https://miam.limsi.fr/
4 Thériaque database: http://www.theriaque.org/
5 OBO Foundry principles: http://obofoundry.org/principles/fp-000-summary.html
September 2020
describe potential drug interactions, although this is the main type of interactions that are
generally described in the literature. Additionally, DIDEO is extended to cover interac-
tions between drugs and natural products including vitamin, mineral, or herbal supple-
ments [20].
3. REQUIREMENTS
Ontology requirements are commonly expressed through competency questions [13],
which are natural language questions that illustrate typical knowledge required from the
ontology. The following competency questions guide the design of FIDEO:
1. What foods potentially interact with simvastatin?
2. Which drugs potentially interact with grapefruit juice?
3. Which cardiovascular drugs may interact with grapefruit juice?
4. What type of interaction mechanisms underlie the interaction between grapefruit
juice and simvastatin?
5. What type of studies describe the interaction between grapefruit juice and sim-
vastatin?
6. What is the level of clinical importance of the grapefruit - simvastatin interaction?
7. Which citrus fruits can be safely consumed by patients taking simvastatin?
8. What alternative drugs can be taken to avoid the interactions between simvastatin
and grapefruit?
4. MATERIALS
Three main sources of information are consulted to identify knowledge to be represented
in FIDEO: a drug interaction compendium, existing corpora of scientific publications
about food-drug interactions, and the contents of the Thériaque database. First, we con-
sult a corpus of relevant scientific publications retrieved from the MEDLINE database.
Previously, the POMELO corpus [15] related to food-drug interactions has been con-
structed using a query in the MEDLINE database with the following terms ([MH] indi-
cates that the keyword had to appear among the MeSH terms chosen to index the arti-
cles):
(‘‘FOOD DRUG INTERACTIONS’’[MH] OR ‘‘FOOD DRUG INTERACTIONS*’’) AND
(‘‘adverse effects*’’)
However, a bibliographic analysis of references cited in a compendium on drug interac-
tion information (i.e., the 8th edition of the Stockley’s Drug Interactions6 ) shows that the
POMELO corpus only covers 3% of scientific articles related to food-drug interactions
that are cited in the Stockley [5]. As this reference is widely used by pharmacovigilance
professionals from France, we additionally make use of a larger corpus that contains
1610 abstracts including articles cited in the Stockley compendium. Finally, we consider
the types of interaction mechanisms used in Thériaque. Although this list is not meant to
be exhaustive, it has the added advantage that it is focused on the most well understood
interaction mechanisms.
6 Stockley: https://about.medicinescomplete.com/publication/
stockleys-drug-interactions/
September 2020
5. METHODOLOGY
We present the conception of FIDEO according to the METHONTOLOGY [10] method-
ology. The first version of the FIDEO ontology follows a waterfall-like process but this
knowledge domain can only be accurately represented by an ever-evolving ontology,
therefore a more iterative way of maintaining the ontology will be considered for future
versions of the ontology [1].
Specification The FIDEO ontology represents knowledge that is necessary to describe
interactions between foods and drugs. This ontology makes it possible to structure in-
formation related to these interactions, facilitating its exploitation within Thériaque. In
addition, for interactions that involve a drug class or a food class, it is useful to infer
that potential interactions may occur if other drugs or foods from the same class are in-
gested concomitantly. Finally, FIDEO can be used to identify food-drug interactions in
future scientific articles for feeding the Thériaque database when new knowledge about
this type of interactions arises. Users of FIDEO have different profiles: (i) curators of
Thériaque interested in integrating FIDEO in the search process of their search engine,
(ii) end users of Thériaque’s search engine, including health professionals and patients,
may browse FIDEO, and (iii) researchers in Natural Language Processing (NLP) may
use it as background knowledge to advance research in information extraction methods
from medical text. Relevant research areas include food- and drug-related entity extrac-
tion, extraction of adverse drug effects, and relation extraction.
Figure 1. Representation of potential food-drug interactions in FIDEO. Prefixes show concept provenance.
Knowledge acquisition Existing ontologies are investigated using BioPortal7 , a reposi-
tory that contains the highest number of biomedical ontologies. Starting from the main
entity types of interest within FIDEO, we identify relevant ontologies, ontology design
7 BioPortal: https://bioportal.bioontology.org
September 2020
patterns and external entities to be reused. We first search for ontologies describing
drug interactions because pharmacology experts we consulted report they are very simi-
lar to food-drug interactions. The main ontologies describing drug interactions are DIO
(Drug Interaction Ontology) [27], DINTO (Drug-Drug Interactions Ontology) [17] and
DIDEO (Drug-drug Interaction and Drug-drug Interaction Evidence Ontology) [7], all
three aligned with the upper ontology BFO [11]. Next, we study existing ontologies re-
lated to foods. When analyzing the interactions mentioned in our corpus, we find that in
addition to the foods themselves and their categories (e.g., cruciferous vegetables, foods
containing tyramine), it is also necessary to represent their cooking preparation and/or
preservation methods since these methods may be involved in food-drug interactions
(e.g., grilled meat, infused tea, frozen grapefruit juice).
When searching for these terms in BioPortal, we find that a Food concept exists
in ChEBI (Chemical Entities of Biological Interest) [16], which is particularly inter-
esting since this ontology is used in DIDEO to represent chemical substances. How-
ever, this concept is defined as a role according to BFO (role being a descendant
of specifically dependent continuant) while chemical substances (which are
used to describe drugs) are material entities (material entity being a descendant of
independent continuant). As we aim to represent these two entities in the same way,
the Food concept of ChEBI is not appropriate on its own. We finally opt for FoodOn
(Food Ontology) [9] because it meets our requirements, with a representation of foods
as material entities as well as the presence of the Food transformation process
concept and its two child concepts Food cooking process and Food preservation
process. Finally, the pharmacology experts involved in the process of designing FIDEO
pointed out that DIDEO does not describe types of interactions but this type of informa-
tion should be represented in the ontology. On the other hand, DINTO contains a DDI
mechanism (Drug-Drug Interaction mechanism) concept and relevant sub-concepts, but
the coverage of interaction types is incomplete. We are currently working on this part in
order to enrich the representation of DINTO interactions, notably according to the Inter-
action Network Ontology (INO) [21] and the interaction types described in the Thériaque
database. INO represents general and species-neutral types of interactions and interaction
networks, and their related elements and relations.
Figure 2. Organisation of top-level concepts in FIDEO according to the ontology they come from.
September 2020
Conceptualisation Following previous work on formally representing drug interactions,
FIDEO makes the distinction between potential interactions as described in biomedical
literature and actual interactions that are typically defined in relation with a specific pa-
tient. Figure 1 shows on the left side information derived from scientific articles (Data)
and on the right side information about biological processes (Real world). FIDEO defines
two core concepts, the concept Potential food-drug interaction that is linked
through a hasPart relationship to the Precipitant food information entity, identi-
fying the food item that causes the relation. We reuse the Object drug information
concept from DIDEO to represent the drug that is impacted by the interaction. Concepts
from the Information Artifact Ontology (IAO)8 are used to denote the link to a scien-
tific publication, that is Data item and Information content entity, with FIDEO
interactions being defined as a subclass of the latter one. On the side of biological pro-
cesses, there are the FIDEO-defined Food product concept and the Drug product
concept from the Drug Ontology (DRON)9 . Other concepts represented here include
those related to the mechanism of interaction described in the Gene Ontology (GO)10
and the assays described in the scientific publication using the Ontology of Biomedical
Investigations (OBI)11 .
Integration Figure 2 illustrates the way top-level concepts are interrelated within FIDEO
and how they are associated with BFO concepts. At the top of the figure, we find the main
concepts referred from BFO, while the bottom of the figure represents concepts defined
in FoodOn. The right hand side of the figure presents the concepts reused from ChEBI.
The center of the image describes the main entities defined by FIDEO creators. The
integration step is performed manually in order to better structure the concepts coming
from the reused ontologies and the new concepts introduced in FIDEO. Thus, drugs are
modelled in Figure 2 through the Chemical substance concept from ChEBI, which
is linked to the Drug product concept of DIDEO according to a partOf relationship.
Drugs and foods are thus represented as material entities (being subClassOf Material
entity in BFO).
For simplicity, we introduce a Food product concept being equivalent to the
Foodon product type concept. This concept has been linked to the ChEBI Food con-
cept via an hasRole relationship for reflecting that a food as a material entity can only
interact with a drug if it is actually used in its nutritional role (i.e., consumed). Food
components, designated as Food component product, are linked to Food product
through a partOf relationship. The concept Chemical substance is also linked to
Food product by a partOf relationship because some foods may contain chemical sub-
stances. This link may be useful in inferring a potential interaction with a food con-
taining a chemical substance that is described as interacting with another chemical sub-
stance in the frame of a drug-drug interaction. The Food transformation process
concept is used to describe the transformation processes that foods can undergo (via an
outputOf relationship). Finally, the Interaction mechanism concept, initially defined
as a DINTO concept, is renamed since its hierarchy will be enriched and modified to
represent the interactions between a food and a drug. This concept is defined as a BFO
8 IAO: https://github.com/information-artifact-ontology/IAO/
9 DRON: https://bitbucket.org/uamsdbmi/dron/src/master/
10 GO: http://geneontology.org/
11 OBI: http://obi-ontology.org/
September 2020
Figure 3. Example of instantiation of a food-drug interactions in FIDEO.
process in FIDEO, similar to other interactions represented in INO. The Food product
and Chemical substance concepts are linked to the Interaction mechanism by an
isAbout relationship.
Implementation FIDEO is implemented in OWL12 and the definition of FIDEO top-
level concepts is done manually using Protégé13 . Importing concepts from different on-
tologies using separate modules is also done manually because we include only drugs
and foods involved in at least one interaction mentioned in our corpus. This choice is
motivated by the fact that ontologies such as ChEBI and FoodOn contain many concepts
that are not relevant for the purpose of FIDEO and are impractical when loaded in full in
Protégé. Figure 3 provides an instantiation example that represents information given in
the following statement from an article about drug interactions of grapefruit juice [14]:
In vitro experiments confirmed that furanocoumarins from grapefruit juice are
both competitive and mechanism-based inhibitors of CYP3A4.
Based on this statement, a new record is added in FIDEO as shown in the figure,
representing evidence from an in vitro experiment about active chemical substances con-
tained in grapefruit juice. These chemical substances modify the usual metabolic process
of the drug through enzyme inhibition. For the first version of the ontology, we use 20
abstracts annotated by drug safety professionals that curate Thériaque and the Bordeaux
pharmacovigilance center within the MIAM project. To map terms used in scientific ar-
ticles to concepts in FIDEO, terms are first normalised in a basic way (e.g., removal
of plurals and special characters, conversion to lower case) and synonyms coming from
the UMLS [2] are recovered. In addition, we consider concatenating terms with several
frequently used phrases including food product, ratio, and of material. To find poten-
tial matches for annotated entities in the reused ontologies, a knowledge graph platform
called Stardog14 is used. This tool allows, among other things, to store large ontologies,
12 OWL: https://www.w3.org/OWL/
13 Protégé: https://protege.stanford.edu
14 Stardog: https://www.stardog.com/
September 2020
Drugs Drug classes Foods Food categories Interactions Interaction mechanisms
134 4 76 2 569 18
Table 1. Statistics about food-drug interactions described in the current version of FIDEO
to perform SPARQL queries and path queries. Each annotated entity is then searched in
the ontologies loaded in Stardog. If a match is found, the concept is imported in FIDEO
and if no match is found, the concept is integrated in FIDEO by hand. Additionally, we
populate FIDEO with all the foods, drugs, food categories and drug classes that are iden-
tified as potentially interacting in the index of the Stockley compendium. The FoodOn
taxonomy is used to link food items to corresponding food categories and the ChEBI
taxonomy is similarly used to add drugs in the appropriate drug class.
Evaluation As FIDEO has not yet been finalised, the evaluation results are still prelim-
inary. Following the implementation of FIDEO with the 20 abstracts annotated by the
experts, the following coverage was obtained for the main entities:
• 82% of drugs and 90% of drug classes are found in ChEBI,
• 62% of foods and 76% of food components are found in FoodOn.
A manual evaluation showed that annotated drugs could not be found in FIDEO be-
cause the abstracts mention branded names not generic names (e.g., Bonefos instead of
Sodium clodronate), acronyms (e.g., Tcy instead of tetracycline) and composed phrases
that are not represented as such in FIDEO (e.g., diclofenac softgel). There are also in-
complete annotations (e.g., hydroxide gels instead of aluminium hydroxide gels) and a
much larger number of foods is not yet represented in FoodOn including food items from
Japanese cuisine (e.g., natto), generic foods (e.g., milk, juice and coffee), and scientific
names (e.g., Amblygaster, Sardinella). These encouraging results indicate that external
ontologies imported by FIDEO cover the knowledge about drugs involved in food-drug
interactions in an acceptable manner and about foods to a lesser degree. With respect
to the compentency questions used to define the FIDEO requirements in section 3, the
current version of the ontology is currently able to answer the questions 1 to 5, while
questions 6 to 8 will be part of future work. This is because at the moment our efforts are
focused on representing evidence about food-drug interactions. Providing options for the
management of food-drug interactions in a clinical setting is an equally important task
that will be addressed in the future.
Documentation The present article is the first publicly available document describing
the creation of FIDEO.
6. DISCUSSION
An important aspect regarding the types of interactions remains unsolved, that is adjust-
ing the granularity of information which is represented in the ontology. An illustration of
this complexity is the change in pharmacokinetic parameters of drugs due to interactions
(e.g., increase in plasma concentration, decrease in bioavailability). While this type of
information is often discussed in scientific articles, it is typically considered too detailed
to be efficiently used in clinical practice and it is beyond the scope of the current version
of FIDEO. To assess the clinical importance of food-drug interactions, medical profes-
sionals typically take into consideration the level of evidence provided in the literature,
but the FIDEO ontology will be populated in the future by a partially automated process
September 2020
based on information extraction from text, including entity recognition, entity linking
and relation extraction. This process will add a level of uncertainty, as automated text
analysis introduces errors as well. Ongoing work investigates a metric of confidence that
provides an indication of the quality of available evidence about an interaction in combi-
nation with information extraction accuracy. A limitation of existing ontologies on rep-
resenting drug interactions that are extended by FIDEO is that they are not aligned with
broader efforts of modelling interactions as biological processes. Current implementa-
tions introduce ambiguities, for example antagonism has a different meaning when used
to describe a physiological effect antagonism in the context of a pharmacodynamic drug
interaction and when used to describe a type of biochemical binding, where a substance
binds to the same site an agonist would bind to without causing activation of the receptor.
7. CONCLUSIONS
In this work, we proposed an ontology that describes potential food-drug interactions
along with supporting evidence from scientific articles. We extend existing work on rep-
resenting foods, drugs and drug-drug interactions to represent food-drug interactions.
The METHONTOLOGY methodology was followed to represent foods, food compo-
nents and food categories as well as specific interaction mechanisms with drugs. The
resulting ontology was manually populated using several abstracts annotated by pharma-
cology experts in addition to information described in a compendium of drug interac-
tions. Evaluation results show higher coverage of drugs and drug classes than of foods,
food components and food categories in the ontologies extended by FIDEO. Future work
will include an integration of automated processes of information extraction from sci-
entific publications, along with a confidence score that provides a combined indication
of text analysis accuracy and evidence level. We will also propose a closer alignment of
drug interaction ontologies with existing work on representing interaction networks.
Acknowledgements This work was supported by the projects MIAM through grant
ANR-16-CE23-0012 and kANNa through grant H2020 MSCA-IF-217 number 800578.
References
[1] E. Blomqvist, K. Hammar, and V. Presutti. Engineering ontologies with patterns-the extreme design
methodology. Ontology Engineering with Ontology Design Patterns, (25):23–50, 2016.
[2] O. Bodenreider. The unified medical language system (UMLS): integrating biomedical terminology.
Nucleic Acids Research, 32(suppl 1):D267–D270, 01 2004.
[3] G. Bordea, S. Faralli, F. Mougin, P. Buitelaar, and G. Diallo. Evaluation dataset and methodology for
extracting application-specific taxonomies from the Wikipedia knowledge graph. In Proceedings of The
12th Language Resources and Evaluation Conference, pages 2341–2347, Marseille, France, May 2020.
European Language Resources Association.
[4] G. Bordea, T. Randriatsitohaina, F. Mougin, N. Grabar, and T. Hamon. Query selection methods for
automated corpora construction with a use case in food-drug interactions. In Proceedings of the 18th
BioNLP Workshop and Shared Task, pages 115–124, 2019.
[5] G. Bordea, F. Thiessard, T. Hamon, and F. Mougin. Automatic query selection for acquisition and
discovery of food-drug interactions. In International Conference of the Cross-Language Evaluation
Forum for European Languages, pages 115–120. Springer, 2018.
[6] M. Brochhausen, P. E. Empey, J. Schneider, W. R. Hogan, and R. D. Boyce. Adding evidence type rep-
resentation to DIDEO. CEUR Workshop Proceedings, 1747, 2016. 2016 Joint International Conference
on Biological Ontology and BioCreative - Food, Nutrition, Health and Environment for the 9 Billion.
[7] M. Brochhausen, J. Schneider, D. Malone, P. E. Empey, W. R. Hogan, and R. D. Boyce. Towards a
foundational representation of potential drug-drug interaction knowledge. In First International Work-
September 2020
shop on Drug Interaction Knowledge Representation (DIKR-2014) at the International Conference on
Biomedical Ontologies (ICBO 2014), 2014.
[8] A. Dahan and H. Altman. Food–drug interaction: grapefruit juice augments drug bioavailability-
mechanism, extent and relevance. European journal of clinical nutrition, 58(1):1, 2004.
[9] D. M. Dooley, E. J. Griffiths, G. S. Gosal, P. L. Buttigieg, R. Hoehndorf, M. C. Lange, L. M. Schriml,
F. S. L. Brinkman, and W. W. L. Hsiao. FoodOn: a harmonized food ontology to increase global food
traceability, quality control and data integration. Nature Partner Journals Science of Food, 2(1):23, Dec.
2018.
[10] M. Fernandez-Lopez, A. Gomez-Perez, and N. Juristo. METHONTOLOGY: from ontological art to-
wards ontological engineering. In Proceedings of the AAAI97 Spring Symposium, pages 33–40, March
1997.
[11] P. Grenon and B. Smith. SNAP and SPAN: Towards dynamic spatial ontology. Spatial Cognition &
Computation, 4(1):69–104, 2004.
[12] P. Grenon, B. Smith, and L. Goldberg. Biodynamic ontology: applying BFO in the biomedical domain.
Studies in health technology and informatics, pages 20–38, 2004.
[13] M. Grüninger and M. S. Fox. The role of competency questions in enterprise engineering. In Bench-
marking—Theory and practice, pages 22–31. Springer, 1995.
[14] L.-Q. Guo and Y. Yamazoe. Inhibition of cytochrome p450 by furanocoumarins in grapefruit juice and
herbal medicines. Acta Pharmacologica Sinica, 25(2):129, 2004.
[15] T. Hamon, V. Tabanou, F. Mougin, N. Grabar, and F. Thiessard. POMELO: Medline corpus with man-
ually annotated food-drug interactions. In Recent Advances in Natural Language Processing (RANLP),
pages 73–80, 2017.
[16] J. Hastings, G. Owen, A. Dekker, M. Ennis, N. Kale, V. Muthukrishnan, S. Turner, N. Swainston,
P. Mendes, and C. Steinbeck. ChEBI in 2016: Improved services and an expanding collection of metabo-
lites. Nucleic Acids Research, 44(D1):D1214–D1219, January 2016.
[17] M. Herrero-Zazo, J. Hastings, I. Segura-Bedmar, S. Croset, P. Martı́nez, and C. Steinbeck. An ontology
for drug-drug interactions. In Semantic Web Applications and Tools for Healthcare and Life Sciences
(SWAT4LS), volume 1114. CEUR Workshop Proceedings, 2013.
[18] M. Herrero-Zazo, I. Segura-Bedmar, P. Martı́nez, and T. Declerck. The DDI corpus: An annotated
corpus with pharmacological substances and drug–drug interactions. Journal of biomedical informatics,
46(5):914–920, 2013.
[19] M. Husson. Theriaque: independent-drug database for good use of drugs by health practitioners. In
Annales pharmaceutiques francaises, volume 66, pages 268–277, 2008.
[20] J. Judkins, J. Tay-Sontheimer, R. D. Boyce, and M. Brochhausen. Extending the DIDEO ontology to
include entities from the natural product drug interaction domain of discourse. Journal of biomedical
semantics, 9(1):15, 2018.
[21] A. Özgür, J. Hur, and Y. He. The Interaction Network Ontology-supported modeling and mining of
complex interactions represented with multiple keywords in biomedical literature. BioData mining,
9(1):1–17, 2016.
[22] G. Popovski, S. Kochev, B. K. Seljak, and T. Eftimov. FoodIE: a rule-based named-entity recognition
method for food information extraction. In In Proceedings of the 8th International Conference on Pattern
Recognition Applications and Methods, pages 915–922, 2019.
[23] G. Popovski, B. K. Seljak, and T. Eftimov. FoodBase corpus: a new resource of annotated food entities.
Database, 2019, 2019.
[24] I. Segura-Bedmar, P. Martı́nez, and M. H. Zazo. Semeval-2013 task 9: Extraction of drug-drug interac-
tions from biomedical texts (ddiextraction 2013). In Second Joint Conference on Lexical and Compu-
tational Semantics (*SEM), Volume 2: Proceedings of the Seventh International Workshop on Semantic
Evaluation (SemEval 2013), volume 2, pages 341–350, 2013.
[25] B. Smith, M. Ashburner, C. Rosse, J. Bard, W. Bug, W. Ceusters, L. J. Goldberg, K. Eilbeck, A. Ireland,
C. J. Mungall, et al. The OBO Foundry: coordinated evolution of ontologies to support biomedical data
integration. Nature biotechnology, 25(11):1251–1255, 2007.
[26] C. Yang, H. Ambayo, B. De Baets, P. Kolsteren, N. Thanintorn, D. Hawwash, J. Bouwman, A. Bronse-
laer, F. Pattyn, and C. Lachat. An ontology to standardize research output of nutritional epidemiology:
From paper-based standards to linked content. Nutrients, 11(6):1300, 2019.
[27] S. Yoshikawa, K. Satou, and A. Konagaya. Drug Interaction Ontology (DIO) for inferences of possible
drug-drug interactions. Studies in Health Technology and Informatics, 107(Pt 1):454–458, 2004.