Vol. 20 no.
1 2004, pages 120–121
BIOINFORMATICS APPLICATIONS NOTE
GIS: a biomedical text-mining system for
gene information discovery
Jung-Hsien Chiang∗, Hsu-Chun Yu and Huai-Jen Hsu
Department of Computer Science and Information Engineering, National Cheng Kung
University, Tainan 701, Taiwan, ROC
Received on March 7, 2003; revised on June 5, 2003; accepted on June 17, 2003
ABSTRACT
Summary: We present a biomedical text-mining system
focused on four types of gene-related information: biological
functions, associated diseases, related genes and gene–gene
relations. The aim of this system is to provide researchers an
easy-to-use bio-information service that will rapidly survey the
rapidly burgeoning biomedical literature.
Availability: http://iir.csie.ncku.edu.tw/~yuhc/gis/
Contact: jchiang@mail.ncku.edu.tw
INTRODUCTION
In the post-genomic era, a great deal of research is directed
toward the analysis and interpretation of genetic sequences
data. The results of these biological experiments are reported
in text form. This type of information, however, is under-
utilized by biologists and medical researchers because of
the highly unstructured and free-format characteristics of
the published information and because of its overwhelm-
ing volume. Currently, there are some systems that analyze
PubMed (http://www.ncbi.nlm.nih.gov/pubmed/) abstracts
and then provide a value-added bio-information service.
For example, Suiseki (Blaschke et al., 1999, http://www.
pdg.cnb.uam.es/suiseki/) focuses on the extraction and visu-
alization of protein–protein interactions. MedMiner (Tanabe
et al., 1999, http://discover.nci.nih.gov/textmining/filters.
html) takes advantage of GeneCards (Safran et al., 2002,
http://bioinfo.weizmann.ac.il/cards/) as its knowledge source
and offers gene information related to specific keywords.
XplorMed (Perez-Iratxeta et al., 2001, http://www.bork.embl-
heidelberg.de/xplormed/) presents specified information
through interaction with the user. EDGAR (Rindflesch et al.,
2000, http://www-smi.stanford.edu/projects/helix/psb00/)
extracts information about drugs and genes relevant to can-
cer from the biomedical literature. GENIES (Friedman et al.,
2001) extracts and structures information about cellular
pathways from the biomedical literature. The goal of our
research was to design and develop an information system
that can streamline the process of retrieving and analyzing
gene-related information from PubMed abstracts, where the
∗ To whom correspondence should be addressed.
120
DOI: 10.1093/bioinformatics/btg369
information about a gene includes biological functions, asso-
ciated diseases, related genes, and gene–gene relations. The
result has been a considerable reduction in the time and effort
required to survey the literature on genes.
We have developed the GIS (Gene Information System)
to perform high-speed gene information discovery. Our sys-
tem has two modules. Module I, gene information screening,
provides information about biological functions, associated
Downloaded from bioinformatics.oxfordjournals.org by guest on May 1, 2011
diseases and related genes for a queried gene. Module II,
gene–gene relation extraction, extracts the gene–gene rela-
tions described in abstracts and estimates whether the relation
between a pair of genes is positive, cooperative, or negative.
In the following we describe the system architecture of the
two modules.
Module I: gene information screening
In this module, the user can screen the information of
biological functions, associated diseases and related-genes
for each gene in the input gene list [Fig. 1(a) at http://iir.
csie.ncku.edu.tw/~yuhc/gis/figure.htm].
The function of this module is achieved through three
agents. The document retrieval agent is responsible for col-
lecting the medical documents from PubMed. The sentence
selection agent selects the generally important parts of the
abstracts, the title and conclusion section, for processing. If an
abstract does not contain a section labeled ‘CONCLUSION:’,
the system selects the last three sentences. The lexicon ana-
lysis agent is the core of this module. It finds, counts and
indexes, with the domain-specific lexicon, the keywords for
biological functions, associated diseases and related genes
that occur in the abstracts. The domain-specific lexicon con-
tains six parts: biological function lexicon, disease lexicon,
gene lexicon, relational keyword lexicon, negative keyword
lexicon and stopword lexicon. We compile the lexicon from
some on-line dictionaries and suggestion from professional
bio-medical researchers. The lexicon contains synonyms to
handle lexical variation.
Module II: gene–gene relation extraction
In this module, users can acquire information about the
relation between a pair of genes [Fig. 1(b) at http://iir.csie.
Bioinformatics 20(1) © Oxford University Press 2004; all rights reserved.

ncku.edu.tw/~yuhc/gis/figure.htm]. The relations here are
classified into three categories, positive, cooperative and neg-
ative, according to gene function. For example, a gene A can
activate another gene B’s function or expression (a posit-
ive relation); gene A and gene B can bind to a complex or
cooperate with each other (a cooperative relation and example
keywords are ‘associate with’ and ‘is conjugated to’); or gene
A can suppress gene B’s function or expression (a negative
relation).
The function of this module is achieved through four agents:
document retrieval, data preprocess, learning process and
relation prediction. Here we introduce only the kernel learn-
ing process and relation prediction agents. The learning
process agent is responsible for generating sentence expres-
sion patterns from training samples consisting of sentences
describing gene–gene relations. Sentence expression pat-
terns stand for the patterns of wording and term distribution
in describing relations, and they are represented as a vari-
ant of decision tree. This agent operates offline beforehand.
The relation prediction agent judges the relations described
in sentences according to sentence expression patterns and
determines whether the relations are positive, cooperative
or negative. Please visit the website of our system for the
algorithms used in this module.
These two modules can be integrated in the gene informa-
tion discovery process when a medical researcher queries the
information of a specified gene in Module I. It then gets a
related gene for the input gene from the result, and then queries
these two genes’ relation in Module II. After querying gene
information in Module I, the user may find that the query gene
and another gene co-occur in the same sentences frequently.
And they have similar biological functions. At this moment,
the user can proceed to use Module II to query the relation
between these two possibly related genes, and this information
can be used in further study.
System features
The GIS system provides the following features. (1) Finding,
counting and indexing the keywords for biological func-
tions, associated diseases and related genes in the abstracts.
(2) Browsing the sentences and abstracts by clicking the
keywords of biological functions, associated diseases and
GIS: a biomedical text mining system
related genes, instead of showing only a list of titles. (3) Using
colors to mark important keywords, to help the user discover
important information easily and to facilitate understanding.
(4) Editing the domain-specific lexicon of biological func-
tions, diseases and genes to meet the user’s needs, and thus to
expand the lexicon. (5) Filtering the content of the abstracts
and reserving only the conclusions or last three sentences of
the abstract.
We have presented here a biomedical text mining system
that screens gene information and extracts gene–gene relations
described in the text. We extract the information about biolo-
gical functions, associated diseases and related genes through
a domain-specific lexicon. We use three kinds of relations—
positive, cooperative and negative—to represent the relation
between a pair of genes. The extraction performance is 0.840
for precision and 0.767 for recall. In brief, we have designed
and implemented new system architecture for the discovery
of gene information in biomedical texts.
REFERENCES
Downloaded from bioinformatics.oxfordjournals.org by guest on May 1, 2011
Blaschke,C., Andrade,M.A., Ouzounis,C. and Valencia,A. (1999)
Automatic extraction of biological information from scientific
text: protein–protein interactions. In Proceedings of the 7th Inter-
national Conference on Intelligent Systems for Molecular Biology
(ISMB’99), pp. 60–67.
Friedman,C., Kra,P., Yu,H., Krauthammer,M. and Rzhetsky,A.
(2001) GENIS: a natural-language processing system for
the extraction of molecular pathways from journal articles.
Bioinformatics, 17, S74–S82.
Perez-Iratxeta,C., Bork,P. and Andrade,M.A. (2001) XplorMed: a
tool for exploring MEDLINE abstracts. Trends Biochem. Sci., 26,
573–575.
Rindflesch,T.C., Tanabe,L., Weinstein,J.N. and Hunter,L. (2000)
EDGAR: extraction of drugs, genes and relations from the
biomedical literature. Pac. Symp. Biocomput., 5, 517–528.
Safran,M., Solomon,I., Shmueli,O., Lapidot,M., Shen-Orr,S.,
Adato,A., Ben-Dor,U., Esterman,N., Rosen,N., Peter,I. et al.
(2002) GeneCards 2002: towards a complete, object-oriented,
human gene compendium. Bioinformatics, 18, 1542–1543.
Tanabe,L., Scherf,U., Smith,L.H., Lee,J.K., Hunter,L. and
Weinstein,J.N. (1999) MedMiner: an internet text-mining tool
for biomedical information, with application to gene expression
profiling. BioTechniques, 27, 1210–1217.
121