=Paper=
{{Paper
|id=Vol-2042/paper5
|storemode=property
|title=BioKB – Text mining and semantic technologies for the biomedical content discovery
|pdfUrl=https://ceur-ws.org/Vol-2042/paper5.pdf
|volume=Vol-2042
|authors=Maria Biryukov,Valentin Gruès,Venkata Satagopam
|dblpUrl=https://dblp.org/rec/conf/swat4ls/BiryukovGS17
}}
==BioKB – Text mining and semantic technologies for the biomedical content discovery==
https://ceur-ws.org/Vol-2042/paper5.pdf
BioKB - Text Mining and Semantic Technologies
for Biomedical Content Discovery
Maria Biryukov, Valentin Grouès, Venkata Satagopam, and Reinhard Schneider
Luxembourg Centre for Systems Biomedicine, University of Luxembourg,
Luxembourg.
maria.biryukov@uni.lu, valentin.groues@uni.lu, venkata.satagopam@uni.lu,
reinhard.schneider@uni.lu,
WWW home page: https://wwwen.uni.lu/lcsb
Abstract. The ever-increasing number of publicly available biomedical
articles calls for automatic information extraction from digitized publi-
cations. We have implemented a pipeline which, by exploiting text min-
ing and semantic technologies, helps researchers easily access semantic
content of thousands of abstracts and full text articles from PubMed
and Elsevier. The text mining component analyzes the articles content
and extracts relations between a wide variety of concepts, extending the
scope from proteins, chemicals and pathologies to biological processes
and molecular functions. Moreover, the relations are extracted along with
the context which specifies localization of the detected events, precon-
ditions, temporal and logic order, mutual dependency and/or exclusion.
Extracted knowledge is stored in a knowledge base publicly available for
both, human and machine access, via web interface and SPARQL end-
point. To address the data accessibility, reusability and interoperability,
all the extracted relations are standardized using unique resource iden-
tifiers (URIs) and a custom ontology based on Genia ontology.
1 Introduction
Information extraction from biomedical literature is becoming a common prac-
tice due to the huge amounts of available textual data, and technological ma-
turity which allows to gain insight into scientific content. Text analysis evolved
from spotting relevant concepts in the text [18, 19] to co-occurrence statistics [20–
22] and, finally, extraction of complex events which seek to reveal cause-effect
relation between various entities involved in the biomedical processes [23–25].
Some approaches use textual data as the only source for the analysis [26, 27]
while some other combine it with experimental data available from dedicated
databases [28, 29]. Although there have been efforts to harmonize the output
of several named entity recognition systems [30, 31], the wealth of the results
obtained from heterogeneous sources has relatively limited outreach due to lack
of a common language: each system typically comes up with its own nomencla-
ture if any [32]. It is where semantic technologies come into play to become an
integral part of the information extraction process. To increase data reusability
2
and interoperability several solutions have been proposed. PubAnnotation [33],
micro [34] - and nanopublications [35] are important examples of how to repre-
sent extracted knowledge in a standardized format as to be accessible and shared
between machines and human.
Knowledge discovery systems and platforms vary in scope. Many of them
are focused on specific sub-domains. For example, EVEX [23] targets directed
interactions between proteins; DisGeNET [36] explores genetic mechanisms of
diseases, while LimTox [37] searches for toxicity associations of compounds, drugs
and genes with the special interest in liver. Other systems adopt less centered
strategies, trying to cover more aspects involved in biomedical processes. One
such system is PolySearch [38], which searches associations between more than
twenty entity types, exploiting data from medical literature, Wikipedia articles
and 14 databases, among which are UniProt, DrugBank and HMDB. While
leading in scope, Polysearch does not specify association types or directional-
ity, leaving these important pieces of knowledge to be completed by the user.
The BioKB platform1 we introduce here aims to discover cause-effect relations
between multiple entity types and deliver standardized representation of knowl-
edge.
The paper is organized as follows. Section 2 gives an overview of the BioKB
platform. In Section 3 we describe the text mining component. Section 4 focuses
on semantic technologies employed by BioKB. Description of the web interface
follows in Section 5. Section 6 offers a discussion, while conclusions are presented
in Section 7.
2 System Overview
Fig. 1. BioKB platform. The publications, initially stored in Solr, are processed by the
text-mining module. The results are then stored in Virtuoso for dissemination through
a web interface and a SPARQL endpoint.
1
Not to be confused with two other independent systems: http://www.cs.
cmu.edu/~biokb/ and http://www.bioinf.mvm.ed.ac.uk/twiki/bin/view/TWiki/
BioKbPlugin.
3
Systems architecture is illustrated in Figure 1. Publications retrieved from
PubMed and PubMed Central are indexed by a Solr instance; each publication
is processed by the text mining component; results are converted to RDF (N-
Quads) and stored in a triple store. To allow both human and machine access to
the knowledge base, SPARQL endpoint provides machine readable access while
a web application allows users to browse the content of the knowledge base. The
web application is developed in Python 3 using the Flask framework and the
SPARQLWrapper library to query the triple store. We use the vis.js [17] library
to render the bio-medical events as a graph.
3 Text Mining Component
Fig. 2. Main stages of the text mining processing. The sentence detailed in Steps II and
III is PGC-1 mediates this increased GLUT4 expression, in large part, by binding to
and coactivating the muscle-selective transcription factor MEF2C. Triggers are marked
in green, entities in red.
The main steps executed by the text mining component of BioKB are:
i) named entity recognition; ii) syntactic parsing; iii) semantic interpretation
(see Figure 2). They are briefly described in the following subsections.
3.1 Named Entity Recognition
During the Named Entity Recognition (NER) stage, biomedical concepts are
identified in the text. Our choice of a NER engine was driven by two major
requirements: a) capability to identify multiple concept types (bio-entities) to
avoid using and synchronizing multiple NER tools within one pipeline; b) ability
of the engine to map entity name to its unique identifier in a dedicated database.
4
The latter is known as normalization process and is indispensable in order to
ensure database and semantic graph coherence. One of the systems which meets
our criteria is Reflect [1]. Reflect recognizes proteins, chemicals, diseases, tissues,
cell types, GO processes. In Step I of the Figure 2 entities identified by Reflect
are marked in turquoise and grey.
3.2 Trigger Generation
Availability of the trigger dictionary is another prerequisite for semantic analy-
sis. Triggers are words or expressions used to describe a biological process. For
example, mediates, increased, expression, binding and coactivating are examples
of triggers in the phrase in Figure 2. Our trigger dictionary is derived from Genia
annotated corpus [2] which is a collection of PubMed abstracts with the detected
biomolecular events of various types: gene expression, (positive/negative) reg-
ulation, binding, cell process etc. Genia corpus is used also to learn so-called
‘knowledge cues’ which express negative statements and author attitude toward
facts being described, such as hypothesis, uncertainty, etc. Each entry in the
trigger/knowledge cue dictionary is assigned a relative weight calculated based
on positive and negative examples learned from the corpus. During the text anal-
ysis, triggers and knowledge cues are detected as dictionary match; those which
satisfy a pre-set threshold are retained. Since Genia corpus is limited to 2000 ab-
stracts, we try to increase potential coverage of the text mining component and
expand triggers and knowledge cues with synonyms using WordNet [4], which
we access via NLTK [3].
3.3 Syntactic Analysis
With the entities and triggers in place, we can proceed toward syntactic analy-
sis. In order to maximize the probability of identification of “subject-predicate-
object” triples (e.g., “RFLAT-1 activates RATES ”), only the sentences with
at least two entities and one trigger are processed. For syntactic analysis we use
Stanford parser [5] with Stanford dependencies [6]. Step II in Figure 2 shows
dependency graph into which the surface structure of the sentence has been
transformed by the parser. A proven benefit of using dependency parsing in
information extraction task is the ability to map syntactic dependencies onto
semantic roles [7, 24, 8].
3.4 Semantic Interpretation
In order to ensure transfer between syntax to semantics, we opt for the rule-based
approach. It consists of assigning semantic roles to entities which are syntactic
arguments of a trigger. As a result, relations are typed (mostly, the type is in-
herited from the type of their trigger) and, whenever applicable, directed. For
example, direction of a regulatory event is from semantic subject (cause) toward
semantic object (theme). On the contrary, relations of type binding and corre-
lation are naturally not directed. We collect syntactic arguments of the triggers
5
via the depth-first search (DFS) of the sentence graph. The rules are applied on
the ensemble of trigger and its dependencies. For example, syntactic subject of
mediates, PGC-1, is the semantic subject of the regulatory event whose predicate
is mediates. Sometimes nodes are merged in favor of a more straightforward se-
mantic interpretation. Thus, increased and expression are jointly interpreted as
Positive regulation, loosing their individual correspondence to Positive regulation
and Gene expression relation types.
Biomedical processes are subject to rich variety of conditions under which
they could take place. We attempt to account for these by processing information
conveyed by certain lexical and syntactic elements. For example, the main event
in Figure 2, PCG-1 mediates positive regulation of GLUT4, is communicated
along with the description of its mechanism introduced by the adverbial clause
headed by trigger verb binding. By taking this bit of information into account
we can logically order the events described in the sentence: (1) PCG-1 binds
and coactivates MEF2C; (2) GLUT4’s expression is increased (Step III of the
Figure 2).
4 Semantic Web Technologies
The choice of using semantic web technologies for this project was dictated
by two main reasons. First, using an ontology to represent the hierarchy of
relationships offers different level of query granularity. For instance, one can ask
if two entities are connected by a property regulates and be able to retrieve
also results for the property increases because the two properties are linked
by a sub-property relation. Additionally, the ontology and thus the hierarchy of
properties can be updated without having to re-process the publications. Besides
this reasoning capability, using semantic web technologies offers full machine
readable access to the complete knowledge base. Not only can the knowledge
base then be used by third parties directly but it becomes possible to combine
BioKB data with external sources using federated queries.
4.1 BioKB Ontology
We created a simple ontology (Figures 3, 4), to represent the hierarchy of classes
and properties that are used to categorize entities and relations identified by
the text-mining component. This ontology is heavily inspired by the GENIA
ontology. Our decision to allow inferences on sub-relationships resulted in the
need to create a custom ontology. Indeed, in the GENIA ontology, relationships
are represented by classes rather than properties. In the proposed ontology, a
relationship between two bio-medical entities can be directly translated to a
single triple, s p o where s and o are the entities and p is a sub-property of
biokb:bioRelation, the top level property in the BioKB model. We then use the
named graphs feature of Virtuoso to add metadata about this relationship. This
includes information such as creation date, provenance and confidence score.
6
Fig. 3. Classes hierarchy of the BioKB on-
tology
Fig. 4. Properties hierarchy of the BioKB
ontology
4.2 Triple Store
In the current deployment of the platform, a single instance of the open source
edition of Virtuoso 7 hosts the knowledge base and provides the SPARQL end-
point. The server hosting the Virtuoso instance has the following characteristics:
128GB Ram, 8 cores, Hard Drive 500GB 10000 RPM. At the date of this pub-
lication, the size of the database is 22GB for 215 million triples. On top of the
content generated by the text-mining module, the different ontologies mentioned
in Section 4.1 have also been loaded into the triple store. The actual number of
triples constituting the BioKB specific content is about 156 million triples. Those
triples are the result of the processing of more than 800 000 publications. About
10 million events were extracted from approximately 6.5 million sentences.
5 BioKB Web Interface
Besides the SPARQL endpoint, we created a web interface to access the BioKB
content. This web application is publicly and freely available at https://biokb.
lcsb.uni.lu. The home page displays a unique search field providing auto-
complete functionality for all supported bio-medicial entities. Once the user
clicks on an entity, the entity page will be displayed. This page shows a tex-
tual description of the entity, the list of most common co-occurrences for this
entity (as a tag cloud) and two tables with the list of relationships involving this
entity as extracted by the text-mining module. Those incoming and outgoing
relationships are also represented visually as an interactive graph (Figure 5). On
this graph, the central node is the entity corresponding to the current page and
all other nodes and edges represent the most common relationships involving this
entity. For each edge, on mouse over, the label and the number of occurrences
7
Fig. 5. Graph visualization of Asthma. Central node is Asthma. Other nodes with
corresponding edges represent relationships identified by the text-mining component.
Each color correspond to a different entity type (Disease, Genes, etc).
of this relationship are displayed. Each node is clickable and leads to the corre-
sponding entity page. Each edge is also clickable and results in the display of the
relationship details page (Figure 6). This page displays the list of publications
where this relationship was found and the specific sentences. On the entity page,
a download button proposes an export of the result of the SPARQL DESCRIBE
command in RDF/XML and in CSV.
6 Discussion
6.1 Use Cases
The primary goal of our information extraction system and knowledge base
is to help researchers focusing on various types of biomedical data analysis.
We illustrate its functionality with two use cases related to disease network
construction and enrichment.
Chronic obstructive pulmonary disease (COPD): the network verification chal-
lenge. Gathering disease-related factors into a large-scale network became a
common practice. Such networks provide a comprehensive model which helps
to elucidate mechanisms involved in pathological processes. For this network
verification challenge, we used our system ability to provide typed, directed (if
applicable) relations between various concepts. We have scanned the literature
and extracted candidate relations which have been verified by a human expert
and made part of the collaborative community curated network yielded by the
8
Fig. 6. Specific bio-medical event. The page shows a list of publications containing this
specific event.
Challenge [15]. Specifically, we have identified gene/proteins related to the dis-
ease condition, characterizing every time the nature of the relations: up- or
under-regulation; correlation; susceptibility or potential involvement (research
hypothesis), as well as contradictory evidence brought down by various articles.
Parkinson’s Disease map: integration and visualization of disease related data
Similar in flavour, our system is used to extract supporting evidence and/or
suggest new candidates for inclusion to disease maps which is another instance
of disease modeling networks. Parkinson’s Disease map is one such example [16].
Step IV in the Figure 2 shows how GLUT4 was approved and appropriately
integrated in the PD map.
6.2 System Strengths, Limitations and Future Work
Our system is constructed with the goal of detailed knowledge extraction from
textual data, its availability to human and machine. Its strength is the ability to
process abstracts as well as full texts; extract semantic relations between various
concept types and contextualize them in terms of location, conditions, logic and
temporary order. A web interface offers public and free access to the knowledge
base while a SPARQL endpoint offers a machine readable access.
Some aspects of the system will be further developed and there remains room
for change and improvement. First of all, the benchmarking of the system ac-
curacy needs to be performed. From the text mining perspective, it operates on
the sentence level which limits its recall. Although extracted knowledge is nor-
malized with respect to concepts and relations, various nomenclatures are used.
To increase knowledge interoperability we plan to adopt Unified Medical Lan-
guage System (UMLS) which capitalizes on straightforward communication be-
tween various systems processing biomedical and health related data. Currently
9
triple store covers main attributes of the extracted relations, such as subject-
predicate-object while contextual aspects need to be incorporated. Future work
will include enriching the scope of entity types, extending the current web ap-
plication by adding, among other developments, an advanced search feature, a
personalized notification system, a REST web service and some bibliographic
management system to easily cite the publications. BioKB will also have to be
continuously extended by processing more publications.
7 Conclusions
In this paper we described an information extraction system along with the
storage database and web interface in the field of biomedicine. The system em-
ploys text mining and semantic technologies to help discovery and accessibility
of biomedical knowledge. As a proof of concept, we have shown its applicability
to disease network construction and enrichment. Along with the strengths, we
have pointed out the system’s limitations and outlined future work directions.
8 Acknowledgements
This work was partially conducted in the scope of the eTRIKS project that
received funding from the European Union and from the European Federation
of Pharmaceutical Industries and Associations as an IMI JU funded project
(no. 115446). The Reproducible Research Results (R3) team of the Luxembourg
Centre for Systems Biomedicine is acknowledged for support of the project and
for promoting reproducible research.
References
1. Pafilis E., et al. Reflect: augmented browsing for the life scientist. Nat. Biotechnol.,
2009, vol. 27, pp. 508-510.
2. Kim Jin-Dong, Tomoko Ohta, Yuka Tateisi and Jun’ichi Tsujii. GENIA corpus -
a semantically annotated corpus for bio-textmining. Proceedings of the Eleventh
International Conference on Intelligent Systems for Molecular Biology, Brisbane,
Australia, 2003, pp. 180-182.
3. NLTK - Natural Language Toolkit. http://www.nltk.org
4. Princeton University “About WordNet.” Princeton University. 2010. http://
wordnet.princeton.edu
5. The Stanford Parser: A statistical parser. https://nlp.stanford.edu/software/
lex-parser.shtml
6. Stanford Dependencies.
https://nlp.stanford.edu/software/stanford-dependencies.html.
7. David McClosky, Mihai Surdeanu, and Christopher D. Manning. 2011. Event Extrac-
tion as Dependency Parsing. In Proceedings of the Association for Computational
Linguistics - Human Language Technologies 2011 Conference (ACL-HLT 2011),
Main Conference.
10
8. Gunes Erkan, Arzucan Ozgur and Dragomir R. Radev. Extracting Interacting Pro-
tein Pairs and Evidence Sentences by usnig Dependency parsing and Machine Learn-
ing Techniques. In Proceedings of the Second BioCreAtIvE Challenge Workshop -
Critical Assessment of Information Extraction in Molecular Biology, April 23-25,
2007.
9. The Gene Ontology Consortium. Gene ontology: tool for the unification of biology.
Nature Genetics, 25(1), 25–29, 2000.
10. Gremse, M., Chang, A., Schomburg, I., Grote, A., Scheer, M., Ebeling, C., and
Schomburg, D. The BRENDA Tissue Ontology (BTO): The first all-integrating on-
tology of all organisms for enzyme sources. Nucleic Acids Research 39, 2011.
11. Kim S, Thiessen PA, Bolton EE, Chen J, Fu G, Gindulyte A, Han L, He J, He S,
Shoemaker BA, Wang J, Yu B, Zhang J, B. S. The PubChem Project. Nucleic Acids
Research, 2016.
12. Schriml, L. M., Arze, C., Nadendla, S., Chang, Y. W. W., Mazaitis, M., Felix, V.,
. . . Kibbe, W. A. Disease ontology: A backbone for disease semantic integration.
Nucleic Acids Research, 40(D1), 2012.
13. Aken, B. L., Achuthan, P., Akanni, W., Amode, M. R., Bernsdorff, F., Bhai, J.,
. . . Flicek, P. Ensembl 2017. Nucleic Acids Research, 45(D1), D635–D642, 2017.
14. Genia Tagger http://www.nactem.ac.uk/GENIA/tagger/
15. Aishwarya Alex Namasivayam et. al. Community-Reviewed Biological Network
Models for Toxicology and Drug Discovery Applications. Gene Regulation and Sys-
tem Biology, vol 10, pp.51 - 66, 2016.
16. Satagopam Venkata et. al. Integration and Visualization of Translational Medicine
Data for Better Understanding of Human Diseases. Big Data. June 2016, 4(2): 97-
108.
17. Vis.js http://visjs.org
18. Leaman, R. and Gonzalez G. (2008) BANNER: An executable survey of advances
in biomedical named entity recognition. Pacific Symposium on Biocomputing, pp.
652-663, 2008.
19. ABNER: a Biomedical Named Entity Recognized. http://pages.cs.wisc.edu/
~bsettles/abner/
20. Li J, Zhu X, Chen JY. Building disease-specific drug-protein connectivity maps
from molecular interaction networks and PubMed abstracts. PLoS Computational
Biology 5(7), 2009.
21. Rosario B and M.A. Hearst. Classifying semantic relations in bioscience texts. In
Proceeding of the 42nd Annual Meeting on Association for Computational Linguis-
tics, 2004.
22. Hoffmann R., Krallinger M., Andres E., Tamames J., Blaschke C., and Valencia A.
Text mining for metabolic pathways, signaling cascades, and protein networks. Sci
STKE, 2005.
23. Landeghem S., et. al. Exploring Biomolecular Literature with EVEX: Connecting
Genes through Events, Homology, and Indirect Associations. Advances in Bioinfor-
matics, 2012.
24. Kilicoglu, H., S. Bergler. Syntactic Dependency Based Heuristics for Biological
Event Extraction. In Proceedings of the Workshop on BioNLP: Shared Task, pp.
119-127, 2009.
25. Bjöorne, L. et. al. Extracting Complex Biological Events with Rich Graph-Based
Feature Sets. In Proceedings of the Workshop on BioNLP: Shared Task, pp. 10-18,
2009.
11
26. Gawronska, B. Erlendsson and B. Olsson. Tracking biological relations in texts:
a Referent Grammar based approach. Biomedical Ontologies and Text Processing,
ECCB 2005.
27. Peng Y. et.al. An extended dependency graph for relation extraction in biomedical
text. In Proceedings of the Workshop on Biomedical Natural Language Processing,
pp. 21-30, 2015.
28. Liekens A. et.al. BioGraph: unsupervised biomedical knowledge discovery via auto-
mated hypothesis generation. Genome Biology, 12(6), 2011.
29. STRING: Protein-Protein interaction networks. https://string-db.org
30. Li C., Liakata M. and Rebholz-Schumann D. Biological network extraction from sci-
entific literature: state of the art and challenges. Briefings in Bioinformatics, 15(5),
pp. 856-877. 2014.
31. Rebholz-Schumann, D. et.al. Assessment of NER solutions against the first and
second CALBC Silver Standard Corpus. Journal of Biomedical Semantics, 2 (Suppl
5), 2011.
32. Johnson H., et. al. Corpus Refactoring: a Feasibility Study. Journal of Biomedical
Discovery and Collaboration, 2007.
33. Kim J. and Y. Wang. PubAnnotation - a persistent and sharable corpus and annota-
tion reposit. In Proceedings of the 2012 Workshop on Biomedical Natural Language
Processing (BioNLP 2012), pp. 202–205.
34. Clark, T., et. al. (2014). Micropublications: a semantic model for claims, evidence,
arguments and annotations in biomedical communications.
35. Mons, B. and Velterop, J. (2009). Nano-Publication in the e-Science era. In Pro-
ceedings of the International Semantic Web Conference, 2009.
36. Piñero J. et. al (2016). DisGeNET: a comprehensive platform integrating informa-
tion on human disease-associated genes and variants. Nucleic Acids Research, vol.
45(D1), pp.D833-D839, 2017.
37. Cañada A. et. al. LimTox: a web tool for applied text mining of adverse event and
toxicity associations of compounds, drugs and genes. Nucleic Acics Reseaech, 45(W
1), 2017.
38. Cheng D. et.al. PolySearch: a web-based text mining system for extracting rela-
tionships between human diseases, genes, mutations, drugs and metabolites. Nucleic
Acics Reseaech, 36, 2008.