Text Mining and Assisted Curation for the Biomedical Literature

OntoGene is a research project which aims at pushing the boundaries of text mining for the biomedical literature. Our work focuses on the extraction of semantic relations between specific biological entities (such as genes, proteins, drugs, diseases) from the biomedical scientific literature (e.g. PubMed). Our approach is based upon high-precision robust syntactic parsing of the target documents, combined with advanced machine learning techniques. Additionally, we provide an environment for assisted curation (ODIN), as an example of a real-world application of biomedical text mining.

We consider community-run evaluation challenges as the best way to provide an independent and unbiased evaluation of text mining tools. We participated in the BioCreative challenges (since 2006), the BioNLP shared task (2009) and CALBC (2010). In BioCreative II  (2006) we obtained very good results in the tasks of detecting protein interactions, and the best results in detecting experimental methods [2]. In BioCreative II.5 (2009) we obtained the best results in the PPI task (extraction of protein-protein interactions) [4]. In BioCreative III (2010) we were among the three best ranked teams in the PPI-ACT task (binary decision whether a paper contains a curatable protein-protein interaction) and PPI-IMT task (detection of experimental methods). In the latter task, we were the only group which managed to return results with full recall (while maintaining near-best AUC score). Our ODIN tool received favourable comments from curators in the IAT (interactive) task. In the "triage task" of BioCreative 2012 we obtained once again the best overall results.

Our selected publications provide a good overview of the techniques used in the OntoGene system. We are based at the Institute of Computational Linguistics (Department of Computer Science) of the University of Zurich

NEWS!

  • July 2014: A highlight proposal by Dr. Fabio Rinaldi at ECCB 2014 has been selected for presentation (10 proposals selected out of 55!).
  • June 2014: We our revising the OntoGene webservices, in order to provide accurate text mining capabilities via remote access through RESTful interfaces.
  • May 2014: Our collaboration with the Wilbur group at the National Center for Biotechnology Information / National Library of Medicine (NLM/NCBI) on BioC continues successfully. A joint paper on BioC implementations has been published. BioC is a new standard for text representation in biomedical text mining.
  • Apr 2014: On-going collaboration with the RegulonDB group: 
  • Mar 2014: Dr. Fabio Rinaldi invited to co-chair SMBM 2014, October 6-7, University of Aveiro, Portugal.
  • Feb 2014: Dr. Fabio Rinaldi gives a set of invited talks at:
  • Jan 2014: OntoGene enters a new strategic partnership with the Data Science group at Hoffmann-La Roche.

PAST HIGHLIGHTS

  • 2013
    • OntoGene's participation in BioCreative 2013 was a resounding success. We were involved in 3 tasks (in task 5 with two slightly different versions of ODIN), obtaining top ranked results in the CTD tasks and very good comments on the BioC and IAT tasks (which did not have a quantitative evaluation). We had a total of 4 papers, 3 presentations and one demo (we are probably the most active participant in BioCreative)! More information and papers here.
    • As part of our 2013 BioCreative participation we provide:
      • PyBioC: a native python implementation of the BioC core
      • ODIN-CTD: a version of ODIN (our curation interface) customized for CTD
    • OntoGene achieves again top-ranked results in a BioCreative task! (task 3 of BioCreative 2013). Information about the OntoGene web services can be found at:
    • Dr. Fabio Rinaldi invited to co-chair LBM 2013, December 12-13, University of Tokyo, Japan.
  • 2012
    • We organized the 5th International Symposium on Semantic Mining in Biomedicine (SMBM)
    • We obtained the best overall results in the 'Triage' task of BioCreative 2012 (in particular due to very accurate entity recognition), as described in this paper
  • 2011
  • 2010
    • In the BioCreative III shared task we were one of only two groups which participated in all of the tasks, obtaining satisfactory results in all of them. We were involved as co-authors in four journal papers in the special issue of BMC Bioinformatics on BioCreative III.
    • In the CALBC competition (2009/2010), which targets large-scale entity detection across the biomedical literature, our system achieved best results in the 'species' and 'diseases' categories.
  • 2009
    • Our participation in the BioCreative II.5 competitive evaluation (2009) of biomedical text mining systems resulted in the best run for the detection of protein-protein interactions (according to the 'raw' AUC iP/R metric). Our system was overall considered as one of the best three.

    FUNDING

    • From Aug 2010 to Aug 2014 we were funded by the Swiss National Science Foundation through the SASEBio project (Semi-Automated Semantic Enrichment of the Biomedical Literature). Additional funding provided by Novartis and Roche.
    • Additionally in 2012-2013 a post-doc in our group (Gintarė Grigonytė) was funded by a Sciex fellowship (see BioTermEvo project).
    • In 2008/2009 we have been funded by the  SNF project Detection of Biological Interactions from Biomedical Literature (grant 100014-118396/1).  

    Notes

    OntoGene is a non-commercial research project. We have nothing to do with a commercial product of the same name. We are also not related to the so-called "Ontogene network".