BROGUE: A Platform for Constructing and Visualizing

“Gene-Mutation-Disease” Relation Knowledge Graphs to Support

Biomedical Research and Clinical Decisions

Dongsheng Zhao

, Fan Tong

, Zheheng Luo

, Sheng Liu

and Wei Song

Information Center, Academy of Military Medical Sciences, Beijing, China

Information Department, No. 920 Hospital of PLA, Yunnan, China

Beijing MedPeer Information Technology Co., Ltd., Beijing, China

Keywords: Gene, Mutation, Disease, Knowledge Graph, Platform.

Abstract: In the era of precision medicine, clinicians need intensive and comprehensive evidence to conduct research

and make decisions. However, current knowledge bases are isolated and lack integration with information

from other databases or literature, constituting an obstacle for clinicians to locate and understand their

interested relations. In this paper, we design a platform development methodology to construct and visualize

a biomedical knowledge graph combining text mining tools and knowledge fusion models with web interface

libraries. The platform thereby provides the functions of knowledge acquisition, integration, storage, search

and visualization, where each concept in the relation is described by its properties, each relation in the database

is located to sentences and each paragraph in the article is translated into Chinese. To further validate the

feasibility and practicability, we applied the methodology to the “gene-mutation-disease” field and built a

Biomedical Relation of Gene-mUtation-diseasE (BROGUE) platform. The platform included 590 high-

quality gene-mutation-disease relations covering a wide range of commonly-used gene (286), mutation (525)

and disease (347) concepts by October 2019. Two tests demonstrated that BROGUE has potential to be useful

for supporting biomedical research and clinical decisions. The platform has been deployed and is publicly

available at http://brogue.medmdt.net/.

1 INTRODUCTION

Over more than two decades, evidence-based

medicine has rightfully become part of the fabric of

modern clinical practice and has contributed to many

advances in healthcare (McCartney et al., 2016),

eventually developing into several reliable systems,

including ClinicalKey (Huslig and Vardell, 2015),

UpToDate (Fox and Moawad, 2003) and medicine

(Meyers, 2000). However, the coming era of

precision medicine poses a challenge to "one size fits

all" evidence-based medicine owing to the latter’s

failure to provide adequate solutions for outliers

(Beckmann and Lew, 2016). Providing more

intensive and comprehensive evidence to the clinician

https://orcid.org/0000-0003-2616-8891

https://orcid.org/0000-0001-6636-8578

https://orcid.org/0000-0003-4516-5901

https://orcid.org/0000-0002-1054-6440

https://orcid.org/0000-0002-4596-5303

is critical for conducting scientific biomedical

research and making a timely clinical decision.

According to the National Research Council

(U.S.), precision medicine is defined as precisely

tailoring therapies to subcategories of disease on the

basis of genomics including genetic, biomarker,

phenotypic, or psychosocial characteristics (Ashley,

2015; Jameson and Longo, 2015). Hence, “gene-

mutation-disease” relations, as an essential

component of precision medicine, have been the

objects of significant study in recent years and several

high-quality biomedical knowledge bases such as

ClinVar (Landrum et al., 2013), COSMIC (Forbes et

al., 2014) and HGMD (Stenson et al., 2017) have

been built. These knowledge bases provide a large

amount of valuable data that have come to play

Zhao, D., Tong, F., Luo, Z., Liu, S. and Song, W.

BROGUE: A Platform for Constructing and Visualizing “Gene-Mutation-Disease” Relation Knowledge Graphs to Support Biomedical Research and Clinical Decisions.

DOI: 10.5220/0009154600520059

In Proceedings of the 13th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2020) - Volume 3: BIOINFORMATICS, pages 52-59

ISBN: 978-989-758-398-8; ISSN: 2184-4305

critical roles in supporting scientific research and

clinical decision-making.

There are nevertheless three major problems with

the above-mentioned existing knowledge bases:

construction efficiency, integration depth, and

presentation forms. First, to ensure the reliability and

validity of the knowledge, their creators have built

these knowledge bases using dedicated and

meticulous curation from domain experts with

medical training backgrounds and annotation

experience. The construction process is a highly

expensive and time-consuming endeavour, and it

becomes increasingly difficult as biomedical data and

findings rapidly grow. Second, different independent

institutions or organizations introduce each

knowledge base, so that these actors develop

knowledge bases under their own standards and

regulations during the process of construction and

maintenance. The lack of a usable and useful mapping

model creates difficulty bridging the gap between

these knowledge bases and relevant databases (e.g.

terminology and literature), further limiting the

coverage and depth of the knowledge. Third, current

knowledge bases display knowledge items using

tabulation, which involves clearly and directly

enumerating the involved biomedical concepts and

corresponding relation types. While this seems

advantageous from the perspective of relations

themselves, this item-level visualization fails to

present the overall picture of a relation network (i.e.

the connections and interactions between biomedical

concepts), at the same time omitting detailed

descriptions of each relation (e.g. subordinate

properties and supporting evidence).

We therefore designed a platform development

methodology for biomedical knowledge graph

construction as well as visualization, and developed a

platform named BROGUE (Biomedical Relation of

Gene-mUtation-diseasE) for feasibility and

practicability validation. Combining several cutting-

edge techniques and widely used tools, we integrated

relevant information about “gene-mutation-disease”

relations from diverse data sources. We provided this

comprehensive knowledge so as to help users locate

and understand their interested entities or relations.

The platform we present in this paper shows great

potential when it comes to inclusive knowledge

integration and sustainable knowledge discovery to

support biomedical research and clinical decisions.

2 METHODS

Figure 1 displays the overall architecture of our

platform. We built this platform on the WAMP Server

with the frontend implemented using HTML, CSS,

JavaScript, and the backend using PHP/Python and

the MySQL database. Functions currently

implemented by our platform include : 1) knowledge

acquisition which obtains relevant information from

distributed and heterogeneous data sources; 2)

knowledge integration which standardizes diverse

data into an intensive and comprehensive knowledge

graph; 3) knowledge storage which organizes linked

information into a structural and relational MySQL

database; 4) knowledge search which retrieves

relations and relation-associated metadata from a

constructed and integrated knowledge graph; and

finally 5) knowledge visualization which presents

relations, relation-related entities and relation-linked

articles with statistics and network graphs.

Figure 1: The overall architecture of BROGUE for

constructing and visualizing “gene-mutation-disease”

relation knowledge graph.

2.1 Knowledge Acquisition

We selected different authoritative databases from

NCBI as data sources of entity, relations and

literature, as Table 1 indicates. Among these

databases, ClinVar provides essential “gene-

mutation-disease” relation items to our platform,

while we introduced Gene (Maglott et al., 2005),

MedGen (Halavi et al., 2018) and PubMed Central

(Roberts, 2001) to assure and improve data integrity

and provenance tracking when it comes to those

relations. After downloading raw data from the NCBI

FTP server, we designed the following rules to filter

BROGUE: A Platform for Constructing and Visualizing “Gene-Mutation-Disease” Relation Knowledge Graphs to Support Biomedical

Research and Clinical Decisions

valid from less valuable data: 1) genes belong to

homo sapiens; 2) mutations cover substitution,

insertion, deletion, and InDel; 3) diseases consist of

abnormality, dysfunction, and syndrome; 4) relations

contain explicit gene, mutation, disease concepts with

at least one literature citation; 5) the ClinVar linking

literatures includes at least two types among gene,

mutation, and disease.

Table 1: Details for different types of data in BROGUE.

Data Type Data Source

Data

Scale

Data

URL

Gene Gene 60,932 1

Mutation ClinVa

222,786 2

Disease MedGen 50,578 3

Relation ClinVa

24,832 2

Literature PubMed Central 24,297 4

1.ftp://ftp.ncbi.nih.gov/gene/DATA/GENE_INFO/Mamm

alia/Homo_sapiens.gene_info.gz

2.ftp://ftp.ncbi.nlm.nih.gov/pub/clinvar/xml/archive/2018/

ClinVarFullRelease_2018-11.xml.gz

3.ftp://ftp.ncbi.nlm.nih.gov/pub/medgen/NAMES.RRF.gz

4.https://www.ncbi.nlm.nih.gov/pmc/

2.2 Knowledge Integration

We selected different authoritative databases centered

on “gene-mutation-disease” relations, then appended

entity properties (mutation type, allele source,

molecular consequence, gene type, and inheritance

type) and literature evidence to extend relation items.

On the one hand, facing inconsistency in entity

nomenclature, we built glossaries including standard

and synonymous mentions for gene, mutation and

disease entities (see Supplementary Information). We

mapped the entities involving the relations to the

concepts in corresponding terminology databases,

thus describing them by the properties of each

concept. On the other hand, instead of being satisfied

with article-level evidence, we constructed a relation

annotation pipeline combining automatic extraction

with expert curation. Under the supervision of the

ClinVar knowledge base, we located and labeled the

candidate entity co-occurrence sentences using text

mining tools, including NLTK (sentence

tokenization) (Loper and Bird, 2002), ezTag (named

entity recognition) (Kwon et al. 2018) and OpenIE

(relation extraction) (Saha, 2018) as well as a relation

mapping model introduced in previous work (Zhao,

Tong and Luo, 2019). Two independent domain

experts with medical training backgrounds

subsequently curated the annotation results, which

another expert would check further in a case of

inconsistent or cyclic judgment. Finally, the curated

results provide detailed sentence-level evidence and

rich context linked to “gene-mutation-disease”

relations.

2.3 Knowledge Storage

Instead of leveraging a graph database storage

solution (Vicknair et al., 2010), we proceeded with a

traditional relational database (MySQL, in this case)

similar to those used by ClinVar, COSMIC, and

HGMD. This is because MySQL meets the demand

of storage capability and query efficiency based on

the entity diversity and relation complexity of our

current “gene-mutation-disease” relation dataset. To

organize dispersed data from distributed and

heterogeneous data sources, we created association

tables for both entities (from terminology databases

to the ClinVar knowledge base) and relations (from

the literature database to the ClinVar knowledge

base) in addition to basic tables, such as dictionary

tables. Once having imported, cleaned and integrated

data from the previous process, our database finally

stores and indexes “gene-mutation-disease” relations

and related entities and articles information.

2.4 Knowledge Search

Once a query containing single entity or multiple

entities is submitted, the platform automatically

transforms non-standard expressions into standard

terminology using a bio-entities thesaurus built in the

previous step, then returns relations involving the

normalized form of searched entities. Unlike

independent gene and mutation concepts, the

intertwined disease notions are formed into a tree

structure. Our platform further extends the original

input disease terms to more specific and detailed

disease names beneath those terms according to the

hyponym hierarchy of MeSH vocabulary.

Simultaneously, the platform records user action to

calculate the searching frequency of each entity and

subsequently presents real-time trending on user

interests and preferences.

2.5 Knowledge Presentation

Not only are simply listed in tables according to the

user’s query, but the database also presents the

retrieved “gene-mutation-disease” relations in

various forms, including comprehensive knowledge

graphs and detailed literature annotations. For the

knowledge graphs, we took advantage of the D3

JavaScript library (King, 2014) to display overall

relation network where entities are treated as nodes

and relations are regarded as edges. In this context,

BIOINFORMATICS 2020 - 11th International Conference on Bioinformatics Models, Methods and Algorithms

the entities of gene, mutation and disease are

designated different colors and described with

different properties. Relation types are labelled on the

edge, which is linked to related literature. For

literature annotation, we designed customized CSS

styles to highlight the sentence-level location of the

relations scattered in the articles. In addition to

labelling the existing relations from the ClinVar

knowledge base, the platform marks novel relations

detected and discovered during the process of

extraction and curation, providing potentially

valuable information for further study. Moreover, we

introduced Google's neural machine translation

system (Wu et al., 2016) to help non-native English

users grasp a better understanding of the context

around the located relation. By implementing

sentences re-tokenization and terminology

substitution, we optimize the English-Chinese

translation results in the biomedical field.

3 RESULTS

3.1 System Description

As Figure 2 illustrates, the graphical interface of our

web-based platform comprises 3 different kinds of

pages， including: 1) Search Page (Figure 2(a)) to

submit query of “gene-mutation-disease” relation;

(a) Relation search page (b) Relation search result page (table)

(e) Linked literature location page (f) Linked literature location page

Figure 2: The graphical user interface of BROGUE “gene-mutation-disease” relation knowledge graph.

BROGUE: A Platform for Constructing and Visualizing “Gene-Mutation-Disease” Relation Knowledge Graphs to Support Biomedical

Research and Clinical Decisions

(a) Top 10 gene concepts of high frequency (b) Top 10 mutation concepts of high frequency

Figure 3: The statistics of entities and relation covered in BROGUE.

2) Relation Presentation Page (Figure 2(b-c)) as the

terminal to enumerate and depict status and property

of “gene-mutation-disease” relation; 3) Literature

Visualization Page (Figure 2(d-f)) as the interface to

provide the location and description of “gene-

mutation-disease” relation. On our platform, users

can easily search and learn about interested entities or

relations through step-by-step operations. These

retrieval results may provide valuable and reliable

information that supports clinical research and

decision-making.

As Figure 3 demonstrates, our platform currently

includes 590 high-quality “gene-mutation-disease”

relations from ClinVar, covering commonly-used

gene (286), mutation (525) and disease (347)

concepts. The platform further linked relations to 527

full-length articles and located in 654 sentences

among these literatures. Beyond that, 25,375 entities

in the relation were described by their properties and

26,594 paragraphs in the article were translated into

Chinese. These were developed into an overall

knowledge graph of “gene-mutation-disease”

relations. 207 novel relations that were not included

in the ClinVar database were also discovered and

annotated in these linking articles, as well as

displayed in a distinct manner. These relations mostly

derive from the literature containing multiple entities

and complex relations, which made it difficult for

experts to extract these relations into ClinVar from

the linking articles based purely on reading and

comprehension.

3.2 Test Use Cases

3.2.1 Genetic Screening

Genetic screening, which mainly refers to

preimplantation genetic screening (PGS), helps lower

the risks of transmitting genetic defects to offspring,

implantation failure, and/or miscarriage during in

vitro fertilization (IVF) cycles (Lu et al., 2016).

Instead of designing and developing multiple probes

or primers for different disease testing, in this case the

clinician can obtain the entire set of potential diseases

on our platform by submitting the mutation calling

results from next-generation sequencing and

bioinformatics analysis.

Hypothetically, a subject is detected as a carrier of

both c.1118C>T and c.1320G>T mutations in the

CDH1 gene. Only after a tremendous effort in

consultation or experimentation can the clinician

begin to understand the pathogenicity of each

mutation in a traditional diagnostic manner. Our

platform, on the contrary, provides a more convenient

and efficient solution. The clinician can easily acquire

an initial impression that CDH1 c.1118C>T is likely

to correlate with hereditary diffuse gastric cancer

while CDH1 c.1320G>T plays a pathogenic role in

BIOINFORMATICS 2020 - 11th International Conference on Bioinformatics Models, Methods and Algorithms

Blepharocheilodontic syndrome (a rare autosomal

dominant condition), simply by conducting a search

for the CDH1 gene on our platform. Variant

properties can further support these results, and the

literature context can confirm them. In this case,

although both mutations are classified into missense

variants which alter amino acid sequences and affect

protein function, CDH1 c.1320G>T is causal to

Blepharocheilodontic syndrome (PMID: 28301459)

while CDH1 c.1118C>T is only weakly associated

with hereditary diffuse gastric cancer (PMID:

28492532). This evidence suggests that it is safe for

the clinician to conclude this subject is, in the future,

more likely to suffer from Blepharocheilodontic

syndrome than they are from hereditary diffuse

gastric cancer.

3.2.2 Differential Diagnosis

It is essential to characterize causative mutation in a

predisposing gene for the sake of differential

diagnosis among various hereditary syndromes with

their own distinctive organ-specific manifestations

that require different surveillance strategies (De Rosa

et al. 2015). Without costly and time-consuming

whole-exome analysis, or even whole-genome

analysis, the clinician can quickly and accurately

locate the pathogenicity mutations for a given disease

and confirm the classification of disease by means of

a few subsequent tailored tests.

For instance, a hypothetical patient was

previously diagnosed with hereditary breast and

ovarian cancer syndrome by traditional testing

methods (transvaginal ultrasonography and

mammography). Classifying the disease further

requires genetic variants information from

bioinformatics analysis. Unlike massive parallel

sequencing, target region sequencing is the purpose

of our platform. Indeed, target region sequencing

takes advantage of prior knowledge. Based on the

MeSH controlled vocabulary, hereditary breast and

ovarian cancer syndromes have four subtypes

including BROVCA1, BROVCA2, BROVCA3, and

BROVCA4. BROVCA1 is associated with 33

pathogenic mutations in BRCA1, while BROVCA2

is correlated with 13 pathogenic mutations in

BRCA2, and BROVCA3 is linked with 1 pathogenic

mutation in RAD51C. Consistent with the relation-

linked literature, the involving gene-targeted

premiers are then designed and applied for PCR. By

obtaining primer sequences, order of markers and

physical distances among D13S260, D13S1699,

D13S1698, D13S1697, D13S171, D13S1695 and

D13S1694 from the Ensembl Genome Browser

(PMID: 23929434), one can only identify and

confirm c.2808_2811del in BRCA2, suggesting that

the patient is most likely to suffer from breast-ovarian

cancer, familial 2 (BROVCA2).

4 DISCUSSION

4.1 Related Work

Instead of concentrating on bridging the gap between

distributed data sources, previous research has paid

more attention to knowledge base construction from

a heterogeneous collection of unstructured, semi-

structured and structured data or relation extraction

from biomedical literature.

We noted earlier that ClinVar, COSMIC, and

HGMD are commonly-used “gene-mutation-disease”

knowledge bases constructed that took months (or

longer) for skilled experts to construct. More

specifically, ClinVar is a submitter-driven repository

that archives reports of relationships among genomic

variants and phenotypes submitted by clinical

laboratories, researchers, clinicians, expert panels,

practice guidelines, and other groups or

organizations. COSMIC is a hand-curated resource

for exploring the effect of somatic mutations in

human cancer; it also offers detailed mutation data

along with additional information such as

environmental factors or patient pre-disposition.

HGMD is a comprehensive core collection of

germline mutations in nuclear genes that underlay

human inherited disease, functioning primarily

through combined electronic and manual search

procedures. Indeed, the constructed knowledge bases

achieve high quality and high acceptance thanks to

tremendous human effort, resources, and work hours,

but they fail to assure provenance tracking of the

curated knowledge, making themselves less

convincing and persuasive.

On the other hand, Singhal et al., Bravo et al. and

Ravikumar et al. worked on “gene-mutation-disease”

relation extraction tasks using text mining algorithms

and tools. Singhal et al. (2016) used a C4.5 decision

tree classifier to extract disease-gene-variant triplets

from all abstracts in PubMed that were related to a set

of important diseases. Bravo et al. (2015) used a

shallow linguistic kernel and dependency kernel to

identify gene-disease relationships from free text in

MEDLINE. Ravikumar et al. (2015) used a

dependency parse graph traversal algorithm to locate

mutation-disease associations within and across

sentences from MEDLINE abstracts. Having

achieved state-of-the-art performance in several

BROGUE: A Platform for Constructing and Visualizing “Gene-Mutation-Disease” Relation Knowledge Graphs to Support Biomedical

Research and Clinical Decisions

evaluations, these systems yet require model

optimization in classifier selection and feature

engineering, as well as knowledge representation and

subsequent knowledge integration like knowledge

fusion.

4.2 Limitations and Future Work

Good performance though our platform has achieved,

there still exists several limitations require further

improvement, which is database integration and

knowledge discovery.

According to a report of 2018 Molecular Biology

Database Collection in the journal Nucleic Acids

Research, there are up to 1,737 databases currently

available and publicly accessible online (Rigden and

Fernández, 2017). Built using an array of diverse

standards and for a number of different purposes,

these databases are difficult to unify without

meticulously designed and thoroughly thought-out

mapping models. We therefore temporarily selected

only one typical and authoritative database for each

element in our system (i.e. genes, mutation, disease,

relation, and literature) according to expert advice.

This lets us prove how our approach is both usable

and useful for platform construction, at the same time

giving us a chance to build a prototype of this

platform for constructing and visualizing “gene-

mutation-disease” relation knowledge graphs. Far

from adequate, the knowledge in our platform

demands supplementation and extension (e.g.

definition and synonym). In the future, we may resort

to crowdsourcing to establish multiple mapping

models for facilitating larger-scale data integration.

Serving a critical role in precision medicine, novel

“gene-mutation-disease” relation discovery from

biomedical literature remains under-studied. This

may be owed to a lack of fine-grained relation types

and high-quality labelled corpus, which together

make it difficult to construct state-of-the-art relation

extraction models. Indeed, even though processed

and curated by experts, the 207 newly-found relations

extracted from 527 articles are far from enough to

support biomedical research and clinical decision-

making in their own right. Recently, numerous deep

learning models have been applied in natural

language processing tasks, achieving strong

performance (Nguyen and Grishman, 2015; Lin et al.,

2016). In future work, we would design a task-

oriented “gene-mutation-disease” relation extraction

model using advanced deep neural networks like the

CNN or attention model, improving the overall

capability of detecting existing associations and

discovering new relations.

5 CONCLUSIONS

In this paper, we designed a platform development

methodology for biomedical knowledge graph

construction as well as visualization. Combining

several cutting-edge techniques and widely-used

tools, we integrated relevant information about

“gene-mutation-disease” relations from diverse data

sources and presented this information in various

manners to help users learn about their interested

entities or relations. Supported by integrated

knowledge and validated by application scenarios, we

have proven our platform capable of supporting

research conduction and decision making, but we

have also shown that it has great potential when it

comes to inclusive knowledge integration and

sustainable knowledge discovery.

ACKNOWLEDGEMENTS

This Research was funded by National Key R&D

Program of China (2016YFC0901900).

REFERENCES

Ashley, E. A. (2015). The precision medicine initiative: a

new national effort. Jama, 313(21), 2119-2120.

Beckmann, J. S., & Lew, D. (2016). Reconciling evidence-

based medicine and precision medicine in the era of big

data: challenges and opportunities. Genome medicine,

8(1), 134.

Bravo, À., Piñero, J., Queralt-Rosinach, N., Rautschka, M.,

& Furlong, L. I. (2015). Extraction of relations between

genes and diseases from text and large-scale data

analysis: implications for translational research. BMC

bioinformatics, 16(1), 55.

De Rosa, M., Pace, U., Rega, D., Costabile, V., Duraturo,

F., Izzo, P., & Delrio, P. (2015). Genetics, diagnosis and

management of colorectal cancer. Oncology reports,

34(3), 1087-1096.

Forbes, S. A., Beare, D., Gunasekaran, P., Leung, K.,

Bindal, N., Boutselakis, H., ... & Kok, C. Y. (2014).

COSMIC: exploring the world's knowledge of somatic

mutations in human cancer. Nucleic acids research,

43(D1), D805-D811.

Fox, G. N., & Moawad, N. S. (2003). UpToDate: a

comprehensive clinical database. Journal of family

practice, 52(9), 706-710.

Halavi, M., Maglott, D., Gorelenkov, V., & Rubinstein, W.

(2018). MedGen. In The NCBI Handbook [Internet].

2nd edition. National Center for Biotechnology

Information (US).

BIOINFORMATICS 2020 - 11th International Conference on Bioinformatics Models, Methods and Algorithms

Huslig, M. A., & Vardell, E. (2015). ClinicalKey 2.0:

Upgrades in a Point-of-Care Search Engine. Medical

reference services quarterly, 34(3), 343-352.

Jameson, J. L., & Longo, D. L. (2015). Precision

medicine—personalized, problematic, and promising.

Obstetrical & gynecological survey, 70(10), 612-614.

King, R. S. (2014). Visual storytelling with D3: an

introduction to data visualization in JavaScript.

Addison-Wesley Professional.

Kwon, D., Kim, S., Wei, C. H., Leaman, R., & Lu, Z.

(2018). ezTag: tagging biomedical concepts via

interactive learning. Nucleic acids research, 46(W1),

W523-W529.

Landrum, M. J., Lee, J. M., Riley, G. R., Jang, W.,

Rubinstein, W. S., Church, D. M., & Maglott, D. R.

(2013). ClinVar: public archive of relationships among

sequence variation and human phenotype. Nucleic

acids research, 42(D1), D980-D985.

Lin, Y., Shen, S., Liu, Z., Luan, H., & Sun, M. (2016,

August). Neural relation extraction with selective

attention over instances. In Proceedings of the 54th

Annual Meeting of the Association for Computational

Linguistics (Volume 1: Long Papers) (pp. 2124-2133).

Loper, E., & Bird, S. (2002, July). NLTK: the Natural

Language Toolkit. In Proceedings of the ACL-02

Workshop on Effective tools and methodologies for

teaching natural language processing and

computational linguistics-Volume 1 (pp. 63-70).

Association for Computational Linguistics.

Lu, L., Lv, B., Huang, K., Xue, Z., Zhu, X., & Fan, G.

(2016). Recent advances in preimplantation genetic

diagnosis and screening. Journal of assisted

reproduction and genetics, 33(9), 1129-1134.

Maglott, D., Ostell, J., Pruitt, K. D., & Tatusova, T. (2005).

Entrez Gene: gene-centered information at NCBI.

Nucleic acids research, 33(suppl_1), D54-D58.

McCartney, M., Treadwell, J., Maskrey, N., & Lehman, R.

(2016). Making evidence based medicine work for

individual patients. Bmj, 353, i2452.

Meyers, A. D. (2000). eMedicine Otolaryngology: an

online textbook for ENT specialists. Ear, nose & throat

journal, 79(4), 268-271.

Nguyen, T. H., & Grishman, R. (2015, June). Relation

extraction: Perspective from convolutional neural

networks. In Proceedings of the 1st Workshop on

Vector Space Modeling for Natural Language

Processing (pp. 39-48).

Ravikumar, K. E., Wagholikar, K. B., Li, D., Kocher, J. P.,

& Liu, H. (2015). Text mining facilitates database

curation-extraction of mutation-disease associations

from Bio-medical literature. BMC bioinformatics,

16(1), 185.

Rigden, D. J., & Fernández, X. M. (2017). The 2018

Nucleic Acids Research database issue and the online

molecular biology database collection. Nucleic acids

research, 46(D1), D1-D7.

Roberts, R. J. (2001). PubMed Central: The GenBank of the

published literature. Proceedings of the National

Academy of Sciences of the United States of America,

98(2), 381.

Saha, S. (2018, August). Open information extraction from

conjunctive sentences. In Proceedings of the 27th

International Conference on Computational Linguistics

(pp. 2288-2299).

Singhal, A., Simmons, M., & Lu, Z. (2016). Text mining

genotype-phenotype relationships from biomedical

literature for database curation and precision medicine.

PLoS computational biology, 12(11), e1005017.

Stenson, P. D., Mort, M., Ball, E. V., Evans, K., Hayden,

M., Heywood, S., ... & Cooper, D. N. (2017). The

Human Gene Mutation Database: towards a

comprehensive repository of inherited mutation data for

medical research, genetic diagnosis and next-

generation sequencing studies. Human genetics,

136(6), 665-677.

Vicknair, C., Macias, M., Zhao, Z., Nan, X., Chen, Y., &

Wilkins, D. (2010, April). A comparison of a graph

database and a relational database: a data provenance

perspective. In Proceedings of the 48th annual

Southeast regional conference (p. 42). ACM.

Wu, Y., Schuster, M., Chen, Z., Le, Q. V., Norouzi, M.,

Macherey, W., ... & Klingner, J. (2016). Google's

neural machine translation system: Bridging the gap

between human and machine translation. arXiv preprint

arXiv:1609.08144.

Zhao, D., Tong, F., & Luo, Z. (2019, February) Construct

Semantic Type of “Gene-mutation-disease” Relation by

Computer-aided Curation from Biomedical Literature.

In Proceedings of 10th International Conference on

Bioinformatics Models, Methods and Algorithms,

BIOINFORMATICS 2019-Part of 12th International

Joint Conference on Biomedical Engineering Systems

and Technologies, BIOSTEC 2019 (pp. 123-130).

BROGUE: A Platform for Constructing and Visualizing “Gene-Mutation-Disease” Relation Knowledge Graphs to Support Biomedical

Research and Clinical Decisions