=Paper=
{{Paper
|id=Vol-1114/Poster_Kobayashi
|storemode=property
|title=PosMed: A Biomedical Entity Prioritisation Tool Based on Statistical Inference over Literature and the Semantic Web
|pdfUrl=https://ceur-ws.org/Vol-1114/Poster_Kobayashi.pdf
|volume=Vol-1114
|dblpUrl=https://dblp.org/rec/conf/swat4ls/KobayashiMIMMDNGTMT13
}}
==PosMed: A Biomedical Entity Prioritisation Tool Based on Statistical Inference over Literature and the Semantic Web==
https://ceur-ws.org/Vol-1114/Poster_Kobayashi.pdf
PosMed: a biomedical entity prioritisation tool
based on statistical inference over literature and
the Semantic Web
Norio Kobayashi1 , Yuko Makita1 , Manabu Ishii1 , Akihiro Matsushima1 ,
Yoshiki Mochizuki1 , Koji Doi1 , Koro Nishikata1 , David Gifford1 ,
Terue Takatsuki2 , Hiroshi Masuya2 , and Tetsuro Toyoda1
1
Integrated Database Unit, Advanced Center for Computing and Communication,
RIKEN, Wako, Japan
{nori,makita,manabu,amatsus,ym,kdoi,koro,gifford,toyoda}@base.riken.jp
2
Technology and Development Unit for Knowledge Base of Mouse Phenotype,
BioResource Center, RIKEN, Tsukuba, Japan
{takatter,hmasuya}@brc.riken.jp
Abstract. Positional MEDLINE (PosMed) is a web application that
quickly prioritises biomedical entities such as genes and diseases based on
statistical significance of associations between these and a user-specified
keyword by employing our original search engine named General and
Rapid Association Study Engine (GRASE). GRASE search is modelled
as an extension of SPARQL search with statistical analysis, which enables
searching over semantic data including not only linked datasets but also
significant extracted semantic links over multiple biomedical documents.
PosMed was originally implemented for in silico positional cloning stud-
ies by prioritizing genes. Further applications include bioresource search
with associated genetic functions or ontologies, and functional interpreta-
tion of gene variants found from exome sequencing of personal genomes.
PosMed is available at http://database.riken.jp/PosMed/.
Keywords: Linked data prioritisation, Statistical search, Text mining,
Omics analysis
1 Introduction
In the life sciences field, a Semantic Web approach that employs machine-
readable linked data prepared from conventional various omics datasets has been
studied to understand biomedical phenomena. However, because the task of gen-
erating semantic links for our biomedical knowledge is too expensive, and such
knowledge is described by a vast amount of human-readable biomedical litera-
ture, this semantic technology is still not widely adopted by biologists.
For practical use of published biomedical data on the Semantic Web, espe-
cially use of data difficult to utilise due to lack of semantic links, it is beneficial
to reinforce acquisition of such data by supplying a hybrid methodology com-
bining not only inferences over that knowledge described as linked data but also
2 PosMed: a biomedical entity prioritisation tool
knowledge supported by statistical significance over a vast number of multiple
raw documents.
Our implementation of this methodology is the search engine named GRASE
[1]. To confirm the problem solving abilities of GRASE for the life sciences,
we developed a simple but effective graphical user interface for GRASE called
PosMed [2] and in 2005 published this service to be accessible by a user’s web
browser. We started with mouse and human gene prioritisation for in silico
positional cloning, and so far extended datasets and the service for intelligent
bioresource search and exome analysis for the next generation sequencing. The
rest of this paper presents a computational model of GRASE search and problem
solving examples using PosMed with our latest datasets.
2 Statistical search model of GRASE
GRASE search is modelled as an extension of SPARQL search with statistical
analysis, which enables searching over semantic data including not only datasets
in Resource Description Framework (RDF) but also significant extracted seman-
tic links over multiple biomedical documents including MEDLINE abstracts.
Direct search (keyword → entity) The GRASE search engine quickly prioritises
biomedical entities such as genes, diseases, drugs and mouse strains based on
statistical significance of associations with a user-specified keyword. More
con-
ab
cretely, for each entity GRASE generates a 2 × 2 contingency table
cd
consisting of the number of documents (a) where both the the keyword and the
entity appear, (b) where the keyword appears but the entity does not appear, (c)
where the keyword does not appear and the entity does appear, and (d) where
neither the keyword nor the entity appear, then applies the Fisher’s exact test
to the contingency table to compute a P-value for the significance of the test.
Inference search (entity → entity) GRASE further infers other entities from
the result of direct search by applying semantic links described by RDF triples
and statistically extracted co-citation relationships over two entities e1 and e2
appearing in a common document
by applying Fisher’s exact test against 2 ×
ab
2 contingency table consisting of the number of documents (a) where
cd
both e1 and e2 appear, (b) where e1 appears but e2 does not appear, (c) where
e1 does not appear but e2 does appear, and (d) where neither e1 nor e2 appear.
Therefore, an entity can be searched via a search path keyword → entity 1 →
entity 2 and its significance computed as 1 − (1 − pd )(1 − pi ), where pd and pi are
P-values of direct search and inference search respectively.
3 Practical applications of PosMed
Since 2005, we have been extending datasets in PosMed to make it possible
to follow an ever-changing trend of biomedical applications. Datasets currently
introduced in PosMed are shown at http://database.riken.jp/PosMed/.
PosMed: a biomedical entity prioritisation tool 3
3.1 In silico positional cloning
A typical application of PosMed is searching genes with user-specified key-
words and chromosomal intervals suggested by linkage analysis. So that inference
searches can be performed such as mouse gene–drug inference and mouse gene–
human gene inference, currently PosMed supports the following datasets:
– up to 352,000 entities including not only genes in mouse, human, rat, Ara-
bidopsis and rice, but also drugs, metabolites, diseases and mouse strains
associated with document sets including up to 9,870,000 documents from
MEDLINE abstracts, OMIM, gene annotation, molecular interaction, Open
Biomedical Ontologies (OBO) [3] including Gene Ontology, Mammalian Phe-
notype, Human Disease Ontology and Plant Ontology, and
– up to 828,000 semantic links including homologue genes and mouse strain–
gene relationships.
In order to realise quick response, the datasets listed above are distributed over
11 computers and these work in parallel.
PosMed was used to prioritise genes in the RIKEN large-scale mouse ENU
mutagenesis project and contributed to successful identification of 65 responsible
genes [4]. PosMed is also used worldwide and successfully narrowed candidate
genes responsible for a specific function after QTL analysis [5].
3.2 Bioresource search in mice and Arabidopsis
One conventional problem for a mouse bioresource database is that knockout
strains are not used when the targeted gene has an unknown function and no
observed phenotype. We introduced 19,885 mouse strains registered in the Inter-
national Mouse Strain Resource (IMSR) [6] to discover wider resources than by
simple keyword search over mouse strain catalogues and this accelerated biore-
source utilisation, especially for those having fewer phenotypic annotations.
PosMed successfully connects these functionally unknown genes to known
genes using molecular interactions, pathway information and co-citations and as
a result enables suggestion of unobserved phenotypic bioresources. PosMed not
only allows users to retrieve mouse bioresources directly with the user’s keywords
described in bioresource annotations, but also inferentially through correspond-
ing documents for mouse and human genes, diseases, drugs, ontologies, pathways,
metabolites, molecular interactions and MEDLINE abstracts.
As an extension to other species, we newly introduced 7,207 Arabidopsis
bioresources and 823 Arabidopsis phenotype observations extracted by human
literature curation, so that PosMed inferentially discovers Arabidopsis biore-
sources as well through correspoonding documents for genes, phenotypes, on-
tologies, co-expressions, molecular interactions and MEDLINE abstracts.
3.3 Functional interpretation of gene variants
PosMed can also be applied to functional interpretation of genetic variants de-
tected by exome sequencing studies using a next generation sequencer. Since
4 PosMed: a biomedical entity prioritisation tool
exome sequencing studies usually find several hundreds or thousands of genetic
variants by comparing samples and controls, prioritisation of the candidate genes
using PosMed is an effective method for further functional analysis.
Users can upload a tab-separated values file with gene IDs and their descrip-
tions. PosMed prioritises genes listed within files by statistical relevance between
the user’s keywords and each gene, and displays ranked genes together with user
uploaded descriptions and associated documents.
4 Discussion and conclusion
We proposed a Semantic Web data search methodology and tool that extends
conventional graph search like SPARQL with statistical text mining over a vast
number of documents. Not only discovering discoveries documents related to
biomedical entities when given a query as does the service GoPubMed, our
PosMed also supports biomedical entity prioritization. Among several software
tools available to prioritise positional candidate genes, PosMed was evaluated as
sepecially highly effective in comparison with two other similar tools GeneSnif-
fer and SUSPECTS [7]. We expect our prioritisation tools to effectively assist
further practical life science studies, making the most of the data extensibility
of the Semantic Web.
References
1. Kobayashi, N., Toyoda, T.: Statistical search on the Semantic Web. Bioinfomatics,
24(7), pp. 1002–1010 (2008)
2. Makita, Y., Kobayashi, N., Yoshida, Y., Doi, K., Mochizuki, Y., Nishikata, K.,
Matsushima, A., Takahashi, S., Ishii, M., Takatsuki, T., Bhatia, R., Khadbaatar,
Z., Watabe, H., Masuya, H., Toyoda, T.: PosMed: Ranking genes and bioresources
based on Semantic Web Association Study. Nucleic Acids Res., 41(Web Server issue),
pp.W109–W114 (2013)
3. Smith, B., Ashburner, M., Rosse, C., Bard, J., Bug, W., Ceusters, W., Goldberg,
L.J., Eilbeck, K., Ireland, A., Mungall, C.J.; OB Consortium, Leontis, N., Rocca-
Serra, P., Ruttenberg, A., Sansone, S.A., Scheuermann, R.H., Shah, N., Whetzel,
P.L., Lewis, S.: The OBO Foundry: coordinated evolution of ontologies to support
biomedical data integration. Nat. Biotechnol., 25(11), pp.1251-1255 (2007)
4. Masuya, H., Yoshikawa, S., Heida, N., Toyoda, T., Wakana, S., Shiroishi, T.:
Phenosite: a web database integrating the mouse phenotyping platform and the ex-
perimental procedures in mice. J. Bioinform. Comput. Biol., 5, pp.1173–1191 (2007)
5. Kato, N., Watanabe, Y., Ohno, Y., Inoue, T., Kanno, Y., Suzuki, H., Okada, H.:
Mapping quantitative trait loci for proteinuria-induced renal collagen deposition.
Kidney Int., 73, pp.1017–1023 (2008)
6. Eppig, J.T., Strivens, M.: Finding a mouse: the International Mouse Strain Resource
(IMSR). Trends Genet., 15(2), pp.81-82 (1999)
7. Thornblad, T., Elliott, K., Jowett, J., Visscher, P.: Prioritization of Positional Can-
didate Genes Using Multiple Web-Based Software Tools. Twin Res. Hum. Genet.,
10(6), pp.861–870 (2007)