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

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

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

4 Scopus citations

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 publicationLinking Literature, Information, and Knowledge for Biology - Workshop of the BioLink Special Interest Group, ISMB/ECCB 2009, Revised Selected Papers
Pages62-70
Number of pages9
DOIs
StatePublished - 2010
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)0302-9743
ISSN (Electronic)1611-3349

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

Keywords

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

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'Learning from positive and unlabeled documents for retrieval of bacterial protein-protein interaction literature'. Together they form a unique fingerprint.

  • Cite this

    Liu, H., Torii, M., Xu, G., Hu, Z., & Goll, J. (2010). Learning from positive and unlabeled documents for retrieval of bacterial protein-protein interaction literature. In Linking Literature, Information, and Knowledge for Biology - Workshop of the BioLink Special Interest Group, ISMB/ECCB 2009, Revised Selected Papers (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