Species identification for gene name normalization

Full text

Page 1

POSTER PRESENTATION Open Access
Species identification for gene name
normalization
Illés Solt1,2*, Domonkos Tikk1,2, Ulf Leser1
From Workshop on Advances in Bio Text Mining
Ghent, Belgium. 10-11 May 2010
Background
Protein interaction networks are expensive to construct
experimentally. Therefore, researchers usually refer to
the literature or domain-specific databases to convey
knowledge on currently known interactions. Yet the task
of manual collection of knowledge from scientific papers
is labor intensive, and therefore should be automated to
the extent possible. For this, an important step is identi-
fying gene and protein names (termed entities). After
identification, gene names must be mapped to database
identifiers to connect them to structured knowledge.
One particular problem in this step are homonymous,
i.e., identical names referring to different genes in differ-
ent species.
Methods
We present different approaches that aim at assigning
species labels to MEDLINE abstracts. We use (1) as a
* Correspondence: solt@informatik.hu-berlin.de
1Knowledge Management in Bioinformatics, Institute for Computer Science,
Humboldt-Universität zu Berlin, Unter den Linden 6, 10099 Berlin, Germany
Full list of author information is available at the end of the article
Table 1 Comparison of methods for document-level species annotation
GS: MeSH termsGS: UniProt references
SpeciesMethodPRFPRF
journal heuristic
SVM
0.908
0.710
0.632
0.775
0.745
0.741
(0.011)
(0.024)
0.231
0.781
(0.021)
(0.046)
Ali Baba [1]
LINNAEUS [2]
0.888
0.900
0.583
0.660
0.703
0.761
(0.033)
(0.030)
0.654
0.659
(0.063)
(0.057) Human
GNAT [3] ( Ali Baba)
GNAT ( LINNAEUS)
0.878
0.609
0.318
0.507
0.467
0.553
(0.056)
(0.037)
0.618
0.944
(0.103)
(0.072)
UniProt0.934 0.0310.060(1.000) (1.000)(1.000)
journal heuristic
SVM
0.146
0.217
0.310
0.289
0.198
0.248
(0.008)
(0.010)
0.468
0.387
(0.015)
(0.019)
Ali Baba
LINNAEUS
0.654
0.665
0.605
0.602
0.628
0.632
(0.031)
(0.032)
0.829
0.838
(0.059)
(0.061)
E.Coli
GNAT ( Ali Baba)
GNAT ( LINNAEUS)
0.771
0.058
0.301
0.415
0.434
0.102
(0.064)
(0.004)
0.730
0.847
(0.118)
(0.008)
UniProt
RegulonDB [4]
0.946
0.857
(0.032)
(0.107)
(0.063)
(0.191)
(1.000)
(0.175)
(1.000)
0.640
(1.000)
(0.275)
Legend: GS - gold standard species labeling. Only human and E. coli shown for brevity. For comparison, we also provide inter-gold standard agreement between
MeSH, UniProt and RegulonDB. Using UniProt as gold standard, only recall can be compared in a cross-corpus sense as UniProt does not reference all papers
mentioning a protein. For the same reason, when using databases for prediction, only precision is comparable.
Solt et al. BMC Bioinformatics 2010, 11(Suppl 5):P5
http://www.biomedcentral.com/1471-2105/11/S5/P5
© 2010 Solt et al; licensee BioMed Central Ltd.

Page 2

baseline, the most frequent species MeSH term of the
corresponding journal represented as MeSH terms; (2)
the prediction of a binary classifier (SVM) for each spe-
cies; (3) species names found by the tools Ali Baba [1]
or LINNAEUS [2]; (4) the species of a normalized pro-
tein mention found by GNAT [3]. For evaluation, we
use two sources as gold standard document-level anno-
tations: The MeSH terms from MEDLINE and the spe-
cies from UniProt and the E. coli-specific RegulonDB
via protein- MEDLINE references.
Results
Measurements on a random set of 200 k abstracts from
MEDLINE are summarized in Table 1. For MeSH term
prediction, the text based methods (Ali Baba, LIN-
NAEUS, GNAT) show stable performance across spe-
cies, while the classification methods, as they rely on
training data, suffer for species with lower prior prob-
ability. For the most frequent species human, the bag-
of-word based SVM overcomes the difficulty of missing
explicit species mention by learning other clues. Using
UniProt as gold standard, learning methods produce
substantially higher recall, indicating that molecular
biology papers are more explicitly mentioning their
focus organisms. There is a considerable disagreement
between gold standard databases, e.g., only 85.7 % of the
papers referenced from a comprehensive E. coli-specific
database are annotated as E. coli by MeSH. Reasons for
this could be, i.e., incompleteness of MeSH annotations
or consideration of orthologs in RegulonDB.
Conclusion
We conclude that there is no one-size-fits-all method
for identifying species in abstracts. For less frequent spe-
cies, direct species mention identification methods work
best. The advantage of using indirect clues could only
be realized for the most frequent species human, sug-
gesting that machine learning methods should be
applied after better balancing the training data. We also
showed that using MeSH term queries to filter papers
poses considerable limitations on recall.
Acknowledgements
Domonkos Tikk was supported by the Alexander-von-Humboldt Foundation.
Author details
1Knowledge Management in Bioinformatics, Institute for Computer Science,
Humboldt-Universität zu Berlin, Unter den Linden 6, 10099 Berlin, Germany.
2Department of Telecommunications and Media Informatics, Budapest
University of Technology and Economics, H-1117 Budapest, Magyar Tudósok
krt 2., Hungary.
Published: 6 October 2010
References
1.Plake C, Schiemann T, Pankalla M, Hakenberg J, Leser U: AliBaba: PubMed
as a graph. Bioinformatics 2006, 22(19):2444-2445.
2. Gerner M, Nenadic G, Bergman C: LINNAEUS: A species name
identification system for biomedical literature. BMC Bioinformatics 2010,
11:85.
3. Hakenberg J, Plake C, Leaman R, Schroeder M, Gonzalez G: Inter-species
normalization of gene mentions with GNAT. Bioinformatics 2008,
24(16):126-132.
4. Salgado H, Santos-Zavaleta A, Gama-Castro S, Peralta-Gil M, Penaloza-
Spinola M, Martinez-Antonio A, Karp P, Collado-Vides J: The comprehensive
updated regulatory network of Escherichia coli K-12. BMC Bioinformatics
2006, 7:5.
doi:10.1186/1471-2105-11-S5-P5
Cite this article as: Solt et al.: Species identification for gene name
normalization. BMC Bioinformatics 2010 11(Suppl 5):P5.
Submit your next manuscript to BioMed Central
and take full advantage of:
• Convenient online submission
• Thorough peer review
• No space constraints or color figure charges
• Immediate publication on acceptance
• Inclusion in PubMed, CAS, Scopus and Google Scholar
• Research which is freely available for redistribution
Submit your manuscript at
www.biomedcentral.com/submit
Solt et al. BMC Bioinformatics 2010, 11(Suppl 5):P5
http://www.biomedcentral.com/1471-2105/11/S5/P5
Page 2 of 2