Learning from positive and unlabeled documents for retrieval of bacterial protein-protein interaction literature

Hongfang D Liu, Manabu Torii, Guixian Xu, Zhangzhi Hu, Johannes Goll

Research output: Chapter in Book/Report/Conference proceedingConference contribution

4 Citations (Scopus)

Abstract

With the advance of high-throughput genomics and proteomics technologies, it becomes critical to mine and curate protein-protein interaction (PPI) networks from biological research literature. Several PPI knowledge bases have been curated by domain experts but they are far from comprehensive. Observing that PPI-relevant documents can be obtained from PPI knowledge bases recording literature evidences and also that a large number of unlabeled documents (mostly negative) are freely available, we investigated learning from positive and unlabeled data (LPU) and developed an automated system for the retrieval of PPI-relevant articles aiming at assisting the curation of a bacterial PPI knowledge base, MPIDB. Two different approaches of obtaining unlabeled documents were used: one based on PubMed MeSH term search and the other based on an existing knowledge base, UniProtKB. We found unlabeled documents obtained from UniProtKB tend to yield better document classifiers for PPI curation purposes. Our study shows that LPU is a possible scenario for the development of an automated system to retrieve PPI-relevant articles, where there is no requirement for extra annotation effort. Selection of machine learning algorithms and that of unlabeled documents would be critical in constructing an effective LPU-based system.

Original languageEnglish (US)
Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Pages62-70
Number of pages9
Volume6004 LNBI
DOIs
StatePublished - 2010
Externally publishedYes
EventWorkshop of the BioLINK Special Interest Group on Linking Literature, Information and Knowledge for Biology, ISMB/ECCB 2009 - Stockholm, Sweden
Duration: Jun 28 2009Jun 29 2009

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume6004 LNBI
ISSN (Print)03029743
ISSN (Electronic)16113349

Other

OtherWorkshop of the BioLINK Special Interest Group on Linking Literature, Information and Knowledge for Biology, ISMB/ECCB 2009
CountrySweden
CityStockholm
Period6/28/096/29/09

Fingerprint

Protein-protein Interaction
Retrieval
Proteins
Knowledge Base
Learning
Bacterial Proteins
Protein Interaction Networks
Proteomics
High Throughput
Genomics
Annotation
Learning Algorithm
Machine Learning
Classifier
Mesh
Tend
Scenarios
Requirements
Term
Learning algorithms

Keywords

  • Document retrieval
  • Learning from positive and unlabeled
  • Protein-protein interaction

ASJC Scopus subject areas

  • Computer Science(all)
  • Theoretical Computer Science

Cite this

Liu, H. D., Torii, M., Xu, G., Hu, Z., & Goll, J. (2010). Learning from positive and unlabeled documents for retrieval of bacterial protein-protein interaction literature. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 6004 LNBI, pp. 62-70). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 6004 LNBI). https://doi.org/10.1007/978-3-642-13131-8_8

Learning from positive and unlabeled documents for retrieval of bacterial protein-protein interaction literature. / Liu, Hongfang D; Torii, Manabu; Xu, Guixian; Hu, Zhangzhi; Goll, Johannes.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 6004 LNBI 2010. p. 62-70 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 6004 LNBI).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Liu, HD, Torii, M, Xu, G, Hu, Z & Goll, J 2010, Learning from positive and unlabeled documents for retrieval of bacterial protein-protein interaction literature. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). vol. 6004 LNBI, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 6004 LNBI, pp. 62-70, Workshop of the BioLINK Special Interest Group on Linking Literature, Information and Knowledge for Biology, ISMB/ECCB 2009, Stockholm, Sweden, 6/28/09. https://doi.org/10.1007/978-3-642-13131-8_8
Liu HD, Torii M, Xu G, Hu Z, Goll J. Learning from positive and unlabeled documents for retrieval of bacterial protein-protein interaction literature. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 6004 LNBI. 2010. p. 62-70. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-642-13131-8_8
Liu, Hongfang D ; Torii, Manabu ; Xu, Guixian ; Hu, Zhangzhi ; Goll, Johannes. / Learning from positive and unlabeled documents for retrieval of bacterial protein-protein interaction literature. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 6004 LNBI 2010. pp. 62-70 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{fc1b05630ad848bbb1fc06e6dd31d6e3,
title = "Learning from positive and unlabeled documents for retrieval of bacterial protein-protein interaction literature",
abstract = "With the advance of high-throughput genomics and proteomics technologies, it becomes critical to mine and curate protein-protein interaction (PPI) networks from biological research literature. Several PPI knowledge bases have been curated by domain experts but they are far from comprehensive. Observing that PPI-relevant documents can be obtained from PPI knowledge bases recording literature evidences and also that a large number of unlabeled documents (mostly negative) are freely available, we investigated learning from positive and unlabeled data (LPU) and developed an automated system for the retrieval of PPI-relevant articles aiming at assisting the curation of a bacterial PPI knowledge base, MPIDB. Two different approaches of obtaining unlabeled documents were used: one based on PubMed MeSH term search and the other based on an existing knowledge base, UniProtKB. We found unlabeled documents obtained from UniProtKB tend to yield better document classifiers for PPI curation purposes. Our study shows that LPU is a possible scenario for the development of an automated system to retrieve PPI-relevant articles, where there is no requirement for extra annotation effort. Selection of machine learning algorithms and that of unlabeled documents would be critical in constructing an effective LPU-based system.",
keywords = "Document retrieval, Learning from positive and unlabeled, Protein-protein interaction",
author = "Liu, {Hongfang D} and Manabu Torii and Guixian Xu and Zhangzhi Hu and Johannes Goll",
year = "2010",
doi = "10.1007/978-3-642-13131-8_8",
language = "English (US)",
isbn = "3642131301",
volume = "6004 LNBI",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
pages = "62--70",
booktitle = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

}

TY - GEN

T1 - Learning from positive and unlabeled documents for retrieval of bacterial protein-protein interaction literature

AU - Liu, Hongfang D

AU - Torii, Manabu

AU - Xu, Guixian

AU - Hu, Zhangzhi

AU - Goll, Johannes

PY - 2010

Y1 - 2010

N2 - With the advance of high-throughput genomics and proteomics technologies, it becomes critical to mine and curate protein-protein interaction (PPI) networks from biological research literature. Several PPI knowledge bases have been curated by domain experts but they are far from comprehensive. Observing that PPI-relevant documents can be obtained from PPI knowledge bases recording literature evidences and also that a large number of unlabeled documents (mostly negative) are freely available, we investigated learning from positive and unlabeled data (LPU) and developed an automated system for the retrieval of PPI-relevant articles aiming at assisting the curation of a bacterial PPI knowledge base, MPIDB. Two different approaches of obtaining unlabeled documents were used: one based on PubMed MeSH term search and the other based on an existing knowledge base, UniProtKB. We found unlabeled documents obtained from UniProtKB tend to yield better document classifiers for PPI curation purposes. Our study shows that LPU is a possible scenario for the development of an automated system to retrieve PPI-relevant articles, where there is no requirement for extra annotation effort. Selection of machine learning algorithms and that of unlabeled documents would be critical in constructing an effective LPU-based system.

AB - With the advance of high-throughput genomics and proteomics technologies, it becomes critical to mine and curate protein-protein interaction (PPI) networks from biological research literature. Several PPI knowledge bases have been curated by domain experts but they are far from comprehensive. Observing that PPI-relevant documents can be obtained from PPI knowledge bases recording literature evidences and also that a large number of unlabeled documents (mostly negative) are freely available, we investigated learning from positive and unlabeled data (LPU) and developed an automated system for the retrieval of PPI-relevant articles aiming at assisting the curation of a bacterial PPI knowledge base, MPIDB. Two different approaches of obtaining unlabeled documents were used: one based on PubMed MeSH term search and the other based on an existing knowledge base, UniProtKB. We found unlabeled documents obtained from UniProtKB tend to yield better document classifiers for PPI curation purposes. Our study shows that LPU is a possible scenario for the development of an automated system to retrieve PPI-relevant articles, where there is no requirement for extra annotation effort. Selection of machine learning algorithms and that of unlabeled documents would be critical in constructing an effective LPU-based system.

KW - Document retrieval

KW - Learning from positive and unlabeled

KW - Protein-protein interaction

UR - http://www.scopus.com/inward/record.url?scp=77953743348&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=77953743348&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-13131-8_8

DO - 10.1007/978-3-642-13131-8_8

M3 - Conference contribution

AN - SCOPUS:77953743348

SN - 3642131301

SN - 9783642131301

VL - 6004 LNBI

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 62

EP - 70

BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

ER -