Text-mining needs of the food microbiology
research community
Estelle Chaix1, Sophie Aubin2, Louise Deléger1, and Claire Nédellec1
1 MaIAGE, INRA, Université Paris-Saclay, 78350 Jouy-en-Josas, France
2 DIST, INRA RD 10, Route de Saint-Cyr, Université Paris-Saclay, 78026 Versailles
CEDEX, France
firstname.lastname@inra.fr
Abstract. To ensure the usefulness of a bioinformatics service, analysis
of user needs is an essential step. Furthermore, if the service anticipates
the identified needs, acceptance by the user is easier. The aim of this
work is to provide an overview of the requirements of a microbial di-
versity research community for ontology-based text-mining applications.
This study is part of the development of the European infrastructure
for text-mining, OpenMinTeD, that targets Biodiversity among other
research fields. The requirement analysis was completed through tar-
geted online surveys, interviews, focus group meetings and workshops.
This work yields to a detailed up-to-date landscape of stakeholders (data
provider, producer and consumer), their potential role and their expec-
tations of general interest with respect to text-mining applications. We
introduce a user-centered approach to focus on microbiologist end-user
functional requirements, including application user interfaces. The result-
ing description of these needs guides OpenMinTeD current development
to design and develop activities within text-mining projects for microbi-
ology community.
Keywords: user need analysis, text-mining, microbial diversity
1 Introduction
The microbial diversity field encompasses all questions related to the variety
and the variability of microorganisms and their ecosystems. It is then related
to most activities in the microbiology field, which include human, plant or ani-
mal health, but also to positive effects of microorganisms, such as plant growth,
bioremediation or food processing. As in every Life Sciences domain, microbial
diversity information is spread among many complementary sources of experi-
mental and curated data and among huge amounts of textual documents, such
as scientific publications, industrial technology reports or medical documents.
Text-mining (TDM) and information extraction has been recognized as an ef-
ficient way to extract and formalize information in Life Sciences so that it can
then be combined with other sources of data. Compared to other Life Sciences
domains such as molecular mechanism study which have been largely targeted
2 E. Chaix et al.
by text-mining research, the microbial diversity domain suffers from a lack of
attention. The increasing need of combining genetics information with ecology
information motivates the development of dedicated TDM solutions for micro-
bial diversity studies [4]. This justifies its choice as a use case domain for the
European TDM infrastructure OpenMinTeD (OMTD). A first critical step is the
analysis of the needs of the microbiology community with respect to diversity
questions.
End-users like biologists rarely interact directly with the TDM tools but
rather consume the results through third party applications. Starting from their
needs helps identifying the other stakeholders - from bioinformaticians to legal
experts - who will build those TDM based solutions for them.
1.1 Biological Context
Microorganisms are defined as microscopic living organisms. They are very abun-
dant, even in places where conditions are extreme (e.g. with high and low pres-
sure, high or low temperatures). Any physical location, from the smallest (e.g.
microscopic) to the largest (e.g. a planet) is a habitat for microorganisms. Micro-
bial diversity research aims to study and describe microbiomes, to understand
microorganism interactions, their ecosystems, how they adapt to their ecosys-
tem [1] and their phylogeny. The recent advances in high throughput molecular
technologies, such as DNA-sequencing and metagenomics have deeply changed
biodiversity research methods. Indeed, biology and bioinformatics projects pro-
duce huge amounts of heterogenous information about the genetic sequences and
the species and strains that have been experimentally identified in a given en-
vironment [8]. The availability of tools for the joint analysis of biotopes, taxa
and genetic sequences is an emerging need which is the focus of this work. The
information on organism biotopes is a crucial knowledge in biodiversity research.
The quantity of microbial biotope descriptions increases with the need for fast
and high-scale comparison among microorganism ways of life. While this is a
significant community need, there is no resource that centralizes and formalizes
the knowledge on microorganism habitats, which makes it hard to look up and
compare information. This motivated the development of habitat classifications
as illustrated by the call for standardized classification of metagenome projects
[10] or the OntoBiotope ontology 3 development.
1.2 Microbiology and Text and Data Mining
Database information and literature information are complementary. However
free text fields in databases as well as scientific papers are expressed in natu-
ral language. The formalization of the descriptions by Text and Data Mining
(TDM) [15] is then a compulsory step for any further processing because of the
high variability of the text. The resulting formal representation should allow for
3 Available on Agroportal website
Text-mining for Microbiologists 3
the automatic analysis of descriptions of microorganism biotopes from different
experiments, which could then be compared at a large scale.
Information content analysis and standardization requires TDM tools to au-
tomatically extract relevant spans of text, and normalize or categorize them with
reference resources such as ontologies.
1.3 User analysis context
This requirement analysis has been brought about by the OpenMinTeD (OMTD)
project. The OMTD project aims to build a text-mining infrastructure for both
application domain users and text-mining experts. Guided by the needs of the
microbiologists who are the targeted end-users of our TDM solutions, content
providers, aggregators, and application developers find themselves in the position
to seek out TDM solutions that can be incorporated into their services. Bioinfor-
matics infrastructures play an important role in the microbiology domain.They
cover a wide range of activities, including information and data management
and processing. These infrastructures aggregate data and build and operate the
applications used by the researchers. As such they are also targeted by the
requirement analysis in order to propose adequate tools and methods. The con-
tent and resource providers, legal experts and socio-economic actors are also
considered here as strategic stakeholders for the adoption and success of TDM
solutions, and so, the user analysis was also performed for them 4. A stakeholder
map helps to understand and to validate the feasibility of a microbial end-user
application.
Beyond the OMTD project, this study has been conducted with the general
purpose of providing the TDM community with an inventory of the microbiology
community expectations that take into account the recent evolution of both the
microbiology and the TDM domains.
2 Approach
The OMTD project targets four different research community domains, Scholarly
Communication, Life Sciences, Social Sciences and Agriculture and Biodiversity
through different example of use-case. This work takes place in the Agriculture
and Biodiversity field. Needs analysis was carried out from September, 2015 to
March, 2016 using various methods: interviews, a plenary session, workshops
and meetings with other TDM partners. The methodology is fully described in
the Requirement Methodology report 5. It follows three phases, briefly sketched
here:
4 For more details on the user analysis for all stakeholders, please read the docu-
ment White paper on OpenMinTed Community Requirements available on the Open-
MinTeD website
5 It is available on the OpenMinTeD website
4 E. Chaix et al.
– Phase 1 (the preliminary phase) identified end-users and a first set of their
needs in terms of TDM services. This phase determined the main scope of
the use-case with respect to the expectations of the end-users. It recorded a
preliminary outlook of the end-users towards issues in relation to TDM such
as access to content, licensing or storage.
– Phase 2 identified the main stakeholders (data provider, producer and con-
sumer) in this use-case, and their interactions. This analysis has been shared
across all OMTD application domains.
– Phase 3 validated the requirements through final interviews with end-users.
Exchanges and discussions with the OMTD partners also allowed the iden-
tification of the final main stakeholder types, which are generalizable to all
TDM use-cases. In parallel, a list of requirements was established to guide the
functional specifications of TDM solutions.
3 Results of the user needs analysis
3.1 Preliminary needs and scope of the study
The first phase allowed us to identify the potential end-user stakeholders and
the scope of our TDM application with a peculiar focus on the food domain,
particularly relevant for our research institute. The potential end-users which
identified are numerous and diverse: research laboratories, agrofood companies,
food safety agencies, agro-industrial technology institutes and research network.
Indeed, these different profiles have different aims. In order to better understand
them, we describe here their typology:
Research laboratories The microorganisms specifically added during the food
process are usually known, but the environmental microorganisms that could
interact with the artificial microbiota and modify its composition, are mostly
unknown. Microorganisms may contribute to desirable properties or defects of
the final food product (due to their phenotype and the growth context). Fun-
damental microbiology research needs a better understanding of the role of mi-
croorganisms, whether pathogenic or with beneficial effects for health.
Agro-industrial Technology Institutes One role of the technology institutes
is to foster the link between fundamental research and companies, to provide
access to research and development results. By dissemination of knowledge and
technology innovation, these institutes provide diagnosis, counseling and sup-
ports.
Agrofood Companies Research in food companies pursues technical innova-
tion such as the development of new food products with increasing economic im-
pact. Industrial important characteristics of the microbial flora are food safety,
organoleptic and nutritional benefits. There is a focus on positive flora: for in-
stance, to produce fermented foods (i.e. yoghurt, cheese), companies use starter
cultures with known microorganism flora added as ingredients. Bacteria also
produce substances that are used for the design of new food products [12].
Food Safety Agencies Food Safety Agencies tasks are health risk monitoring,
expertise, research and supply of reference data on food risk. The evaluation
Text-mining for Microbiologists 5
of the risk of foodborne illness is done through the information feedback from
analytical laboratories. Tracing back pathogenic bacteria biotopes would help
identify their origin (food ingredient, production process or transport) and con-
trol their occurrence.
Research Networks National and international research networks disseminate
information among partners and coordinate information sharing and publication
efforts. A focus is on microbiome description standard, and specifically on ge-
netics and biodiversity.
A meeting with representatives of these targeted communities (about twenty
participants) defined the framework for our study. All the participants, either
researchers, companies or service providers, advocate a formal and unified rep-
resentation of microorganism biotope information from various sources as a first
critical step.
At the end of this preliminary study, we focused on a use-case targeting global
microbial biodiversity expressed in texts written in English. This first phase con-
firmed also the interest in food products, and in particular in dairy products.
The microbial scope of the study centered on any type of microorganism consid-
ered as positive flora (thus excluding pathogens). These first potential end-user
communities may be included in or excluded from the final use-case, based on
the general expected outcomes.
3.2 Stakeholders of TDM solutions applied to Microbiodiversity
The second phase is dedicated to the description of all the stakeholders involved
in the target TDM application beyond the end-users, as well as their interac-
tions. Indeed, preliminary discussions with end-users identified different types of
stakeholders of the TDM solution. Then, focus group meetings, workshops and
interviews with forty persons enriched the initial need description performed
at the microbiologist community level in the first phase. These meetings and
interviews were held with representatives of the identified stakeholders, within
and outside the project consortium: microbiology researchers, risk analyzers,
database developers and infrastructure managers. We classified them into five
main types: (1) the end-users, including researchers and industrial communi-
ties (2) e-infra operators and aggregators, including application developers and
knowledge managers; (3) text-miners and then (4) content providers and (5)
curators, whose role is essential to improve the final results of TDM extraction.
The stakeholders and their interactions are summarized in figure 1. The main
characteristics of these groups are briefly detailed below:
– End-Users are characterized as the text mining service consumers; they may
have a socio-economic interest.
– Curators are domain specialists (from research scientific or industrial com-
munities), who ensure high quality of data. Their role is the cleaning of the
text-mining results and the revision of the domain specific resources to im-
prove the final quality.
6 E. Chaix et al.
– Data and services aggregators - infrastructure providers want to extend the
scope of the services they offer.
– TDM experts, either researchers or engineers, develop tools and services to be
used by the TDM infrastructure
– Content providers provide textual and semantic resources that are processed
by the TDM infrastructure
Stakeholders we identified appear to be common to other use-cases from the
Life Sciences and Agriculture and Biodiversity domains, where e-infrastructure
and data aggregators also play an important role as in all experimental domains.
We also identified legal experts as general stakeholders, that share expertise on
legal aspects such as international copyright or intellectual property laws for
design of TDM solutions. For this work on the microbial diversity domain, we
focus our need and requirement analysis on the end-user group, which expressed
specific needs (red rectangle in figure 1). Needs of other stakeholders are less
specific to this use case, are not be detailed here, but help us to take a step back
from the microbiologist needs.
Fig. 1. Interactions among stakeholders of the Microbial Biodiversity Use Case
3.3 User-centered method for end-users need analysis
Phase 1 targets the sub-community of the positive flora in the food microbiology
domain, which involves all types of end-users, including companies. User need
analysis may be difficult to conduct in domains where user profiles, practices,
experienced difficulties and expectations are very diverse, which is the case in the
Text-mining for Microbiologists 7
microbial diversity domain. In order to deal with this situation and to express
requirements of end-users, we used a user-centered method.
The user-centered methodology revolves around profiling and analyzing users,
in order to identify the specific characteristics of different types of users. This was
carried out through the persona methodology [7]. Personas are not real persons
but fictional representations of a user, which could represent real people during
the design process.
The aim is to determine which type of integrated information and which ap-
plication interface fit the needs of each persona. We further want to detail and
prioritize aims and behaviors of personas regarding TDM solutions. To do this,
we used an approach comparable to Calde et al. [5] to design our personas. Phase
1 and 2 allowed us to roughly sketch the profile of some representative personas,
and we selected four of them. To complete the description of personas, we orga-
nized a dedicated one-day workshop that involved French microbiologists from
Inra, the French National Institute for Agricultural Research. Participants were
split into four groups, one group per persona. In order to guide the formulation
of the description, we provided four questions to be answered by the partici-
pants, adapted from the Lean Canvas [14]: (1) Who is the persona? (name, age,
function, skill) (2) What kind of information is s/he interested in? (3) What kind
of access/needs would be useful? and (4) What solutions would satisfy his/her
needs and those of his/her community? The persona description is provided in
Table 1. The participants in the different groups also designed collaboratively
the wireframes of the application corresponding to each personas needs. Figure 2
presents an example of the access to the TDM results for persona 3 (“Patrice”).
It enables the user to search strains associated to a phenotype extracted by
text-mining, and provides the link to the original publications so that the user
can check the experimental setting and the relevance of the result. The results
on persona and application design were then shared among all groups through
presentations at the end of the day.
Based on the work of Calde et al.[5], we defined our personas as “secondary
personas”, that is, a main application interface will serve the needs of the differ-
ent types of secondary persona with minor modifications/additions .
3.4 From persona to end-users needs
Personas allowed us to identify various specific needs with respect to the food
ecosystem knowledge. We refer to them using their fictional first names to illus-
trate our interpretation (see Table 1 for the detailed needs of persona).
Microbial biodiversity All personas expressed a need related to knowledge on
microorganism identification and provenance, i.e. to study and characterize the
microbial biodiversity of food ecosystems (such as Yann and Claudia), and to
compare food biodiversity to other ecosystems in order to qualify microorganism
provenance (such as Lily and Patrice). Moreover, from the point of view of
traceability, knowledge about the known living places of microorganisms would
contribute to prevent contamination, either in the medical environment or in the
food industry (as expressed by Yann). An extension of this need is a tool that
8 E. Chaix et al.
Table 1. Persona descriptions
Yann, 35 years old, SME company in bio-preserved food product
What information? Information and literature about bio-preservation process to
answer the following questions: what non-pathogenic bacteria
grows in a given type of food and contributes to longer preserve
the food quality? What are the chemical compounds produced
or degraded by the bacteria that might interfere with given food
products?
What needs? An easily accessible database with a simple and user-friendly
search engine, that he can query about the biopreservation of
the type of food products he is interested in.
What solutions? A database of the bacterial strains that contribute to biopreser-
vation linked to the food products of interest and that exhibit
the relevant growth phenotype (e.g. pH, temperature...). The
database should include information about the public or private
institutes that provide the strain, the patent if any, as well as
organoleptic properties and health impact.
Lily, 38 years old, former baker and expert in bread leavens
What information? She wants to compare the functional diversity of commercial
and artisanal leavens. She uses the information from scientific
papers, collections of national and international organisms, lo-
cal expertise and expertise from baker networks
What needs? She needs to know about the habitats of microorganisms used
in bread making, the physico-chemical characteristics of starter
cultures for bread and the phenotypes of the species listed in
leavens.
What solutions? Pooling data obtained on bacteria and yeasts (and their inter-
action), with an analysis of phenotypes (at strain, species and
leaven levels). Using the knowledge about the taxonomic diver-
sity of yeasts together with phenotype information of strains,
species and leaven in order to compare the functional diversity
of commercial and artisanal leavens.
Claudia, 40 years old, bacteria physiology researcher
expert in aroma compounds produced by lactic bacteria
What information? She wants to use new strains for the production of aroma in
new food products. She first needs information about aroma
compounds that bacteria produce and about where these bac-
teria live. She collects samples of cheese or dairy products, and
she traces the technological process of these products: which
milk? Which treatment? Which manufacturing process? She
uses metagenomics experiments (DNA and RNA information
such as RNA 16s) to identify bacteria strains.
What needs? She would need access to a database on the bacterial metage-
nomic and phenotypic data, with non-redundant and non-
contradictory information. She works on little known species,
so she needs a reliable database to identify and compare their
own genetic data to those of the database, and identify where
they come from.
What solutions? Pooling of data from text in a database about bacteria geno-
types and phenotypes that she could query with a bacterial
phenotype using a user-friendly interface.
continued on next page
Text-mining for Microbiologists 9
continued from previous page
Patrice, 55 years old, microbiologist engineer
working in the R&D department of a dairy company
What information? He wants to gather information for the design of new plant
food products obtained by fermentation. He uses scientific lit-
erature (PubMed bibliographic database, articles, journals, and
conference reports) and student reports and he interacts with
technical centers and researchers.
What needs? He needs a tool that allows him to quickly identify and get a
bacterial strain that meets his requirements: for example, milk
and cereal growth medium, degradation of sugar lactose, pro-
duction of the buttery flavor diacetyl molecule.
What solutions? A database that associates strains, phenotypes and growth con-
ditions (e.g. lactose consumption, diacetyl production acidifica-
tion capacity), and the link to publication for more information.
could be used in the food risk analysis community, enabling cross-checking in an
easy manner by safety agencies.
Microbial phenotypes A second set of needs is related to knowledge on mi-
croorganisms. Microorganisms are massively used in food industry, firstly as a
fermentative agent (as expressed by Lily and Patrice) but not only. Their pecu-
liarity is the production of some nutritional or sensory compounds of interest,
such as aromas (as illustrated by Claudia and Yann). Furthermore, a given mi-
croorganism can either have a positive or an undesirable effect, according to the
context in which it produces the molecule. The study of molecules produced by
bacteria gathers a large community of researchers [13].
Industrial uses There is a need to study and characterize the phenotypes
of microorganisms for food quality improvement, e.g. flavor and taste, nutri-
ments, biopreservation or optimization of food engineering processing. Indus-
trial researchers constantly look for new food products to meet specific diets
and special dietary needs (such as Yann and Patrice). A database referencing
knowledge on microbial phenotypes would allow researchers to easily detect a
species that is able to degrade or consume a specific molecule. Biologists want to
know which molecule was produced by which strain, but also in which substrate,
food or habitat the strain can live. The challenge for biologists is to link bacteria,
habitats including artificial media and produced molecules. Without the infor-
mation on growth media, the production of the molecule by the bacteria may
be jeopardized. This information is generally expressed in scientific publications,
magazines, tables, or personal databases (as described by Patrice). In the spe-
cific case of new food processing, the integration of heterogenous data would
enable users to quickly search for useful microorganisms (data extracted from
the literature and existing genomics databases such as detailed by Claudia).
10 E. Chaix et al.
Fig. 2. Application wireframe for the persona “Patrice”, the microbiologist engineer
4 Scenario: from end-users analysis to TDM solutions
Based on this analysis, we designed a microbial use-case scenario to assist mi-
crobiology sub-communities research by TDM. Three main needs were identi-
fied, that is the completion of the current microbial knowledge with respect to
their biodiversity, phenotypes and usability. The scope of the TDM application
was defined as the recognition and normalization of relevant entities, namely
microorganisms, habitats, molecules, applications and phenotypes, in scientific
publications; and the identification of the relationships between these entities.
One specific final application will focus on positive flora, and in particular on
dairy food. To do this, we aim to leverage state-of-the-art tools and resources.
However, until recently, there have been few attempts to apply text-mining to
microorganism biodiversity study. The Bacteria Biotope task of the BioNLP
Shared Task in 2011 [4], 2013 [3] and 2016 [9] aimed at the extraction and
normalization of bacteria and habitats from text, and their linking. Dedicated
text-mining methods, benchmarks to evaluate these methods, and resources were
developed and will be reused for our use-case. A main need is the normalization
of TDM results with knowledge resources (e.g., taxonomies, ontologies) so that
they can be aggregated, integrated and compared according to a same reference.
There are few standard resources for the food domain with respect to microbial
point of view. The OntoBiotope ontology was built for the description and the
normalization of microorganism habitats. Another existing ontology is the En-
vironment Ontology (EnvO), even if the description level appears more relevant
for macroscopic habitats. Indeed, it has been shown that OntoBiotope yields
better results than the general purpose EnvO for microbiology habitats [16] [6].
The EFSA classification FoodEx2 part on food contamination is also a valuable
Text-mining for Microbiologists 11
source. However, these specific resources may need adaptation to fully meet the
needs of microbiologist users.
The automatic extraction of molecules and microbial phenotype descriptions
is a new open research question in the food domain. Previous results obtained
on molecule name extraction (e.g. BioCreative IV CHEMDNER Task) will be
reused [11]. The expressions of phenotypes at various levels (molecular, physio-
logical and application) are more difficult to handle and may require the devel-
opment of a new ontology.
Finally, the TDM scenario includes the extraction of relations between nor-
malized entities, such as a given microorganism produces a given molecule. Su-
pervised machine learning methods trained with manually annotated examples
will be used. Generic methods such as TEES [2] or AlvisRE [17] have demon-
strated their efficiency when applied to relation extraction in biology. The TDM
results will then be aggregated with other heterogenous data in the bioinformat-
ics application, and will be available to the biological end-users.
5 Conclusion
The elicitation of TDM requirements in the food microbiology domain confirms
that there is a mosaic of needs. We identified the main needs from fictional users
and representatives of the sub-community of the food positive flora microbiol-
ogists, that are the targeted end-users. TDM users want to study, characterize
and compare the microbial biodiversity of food ecosystems. Another need is
the discovery of information among large sets of articles. Aggregation of het-
erogenous data and relevant information from various sources is also a strong
requirement. The persona based approach allowed us to design precise scenarii
involving complex TDM workflows, adequate existing knowledge resources, and
relevant textual sources. To answer the end-user needs, we involved intermediate
stakeholders who are currently working on the integration of the TDM results
in their bioinformatics applications. These detailed user needs also allowed us to
participate in the definition of the OpenMinTeD platform concrete requirements
that ground its implementation.
Acknowledgements
This work was supported by the French National Institute for Agricultural Re-
search (Inra) and the OpenMinTeD project (EC/H2020-EINFRA 654021). This
work was carried out with the technical help of Patricia Gerretto (Inra) and
Vivi Katifori (Agroknow). We would like to thank biologists, bioinformaticians
and all persons who have participated in the end-user need analysis; and in par-
ticular the Inra Florilège working group, Food Microbiome project, Biological
Resource Centers, Bioinformatics infrastructure Migale from IFB (French Insti-
tute of Bioinformatics), ANSES (the French Agency for Food, Environmental
and Occupational Health & Safety), JGI (Joint Genome Institute) and GBIF
(Global Information Facility).
12 E. Chaix et al.
References
1. Bent, S.J., Forney, L.J.: The tragedy of the uncommon: understanding limitations
in the analysis of microbial diversity. The ISME journal 2(7), 689–695 (2008)
2. Björne, J., Salakoski, T.: Tees 2.1: Automated annotation scheme learning in the
bionlp 2013 shared task. In: Proceedings of the BioNLP Shared Task 2013 Work-
shop. pp. 16–25. Association for Computational Linguistics (2013)
3. Bossy, R., Golik, W., Ratkovic, Z., Valsamou, D., Bessières, P., Nédellec, C.:
Overview of the gene regulation network and the bacteria biotope tasks in bionlp’13
shared task. BMC bioinformatics 16(10), S1 (2015)
4. Bossy, R., Jourde, J., Manine, A.P., Veber, P., Alphonse, E., Van De Guchte, M.,
Bessières, P., Nédellec, C.: Bionlp shared task-the bacteria track. BMC bioinfor-
matics 13(11), S3 (2012)
5. Calde, S., Goodwin, K., Reimann, R.: SHS Orcas: The first integrated informa-
tion system for long-term healthcare facility management. In: Case Studies of the
CHI2002. pp. 2–16. ACM (2002)
6. Cook, H.V., Pafilis, E., Jensen, L.J.: A dictionary-and rule-based system for iden-
tification of bacteria and habitats in text. ACL 2016 p. 50 (2016)
7. Cooper, A., Reimann, R., Cronin, D., Noessel, C.: About face: the essentials of
interaction design. John Wiley & Sons (2014)
8. Dahllöf, I.: Molecular community analysis of microbial diversity. Current Opinion
in Biotechnology 13(3), 213 – 217 (2002)
9. Deléger, L., Bossy, R., Chaix, E., BA, M., Ferré, A., Bessières, P., Nédellec, C.:
Overview of the bacteria biotope task at BioNLP shared task. In: BioNLP Shared
Task. p. 101. The Association for Computational Linguistics, Berlin, Germany
(Aug 2016)
10. Ivanova, N., Tringe, S.G., Liolios, K., Liu, W.T., Morrison, N., Hugenholtz, P.,
Kyrpides, N.C.: A call for standardized classification of metagenome projects. En-
vironmental microbiology 12(7), 1803–1805 (2010)
11. Krallinger, M., Leitner, F., Rabal, O., Vazquez, M., Oyarzabal, J., Valencia, A.:
Chemdner: The drugs and chemical names extraction challenge. Journal of chem-
informatics 7(1), S1 (2015)
12. Leroy, F., De Vuyst, L.: Fermented food in the context of a healthy diet: how to
produce novel functional foods? Current Opinion in Clinical Nutrition & Metabolic
Care 17(6), 574–581 (2014)
13. Longo, M.A., Sanromán, M.A.: Production of food aroma compounds: microbial
and enzymatic methodologies. Food Technology and Biotechnology 44(3), 335–353
(2006)
14. Maurya, A.: Running lean: iterate from plan A to a plan that works. O’Reilly
Media, Inc. (2012)
15. Pignatelli, M., Moya, A., Tamames, J.: Envdb, a database for describing the envi-
ronmental distribution of prokaryotic taxa. Environmental Microbiology Reports
1(3), 191–197 (2009)
16. Ratkovic, Z., Golik, W., Warnier, P.: Event extraction of bacteria biotopes: a
knowledge-intensive nlp-based approach. BMC bioinformatics 13(11), S8 (2012)
17. Valsamou, D.: Information extraction from scientific articles for the reconstruc-
tion of the biological regulatory network during the seed development phase of
Arabidopsis thaliana. Ph.D. thesis, Université Paris-Saclay, France (2017)