Project

Discipline

text mining; molecular biology; syntactic parsing; semi-automated semantic annotation; information extraction; natural language processing; knowledge management; literature curation

Lead
Lay summary
The understanding of the complex mechanism that support life, and are responsible for health or disease at the cellular level, promises to unveil new technologies that will benefit the health of every citizen, and support the growth of novel industries. The potential of Text Mining applications in the life sciences is widely acknowledged. Scientists and companies are looking for solutions that allow the automated processing of scientific literature. Text Mining tools aim at supporting the process of knowledge gaining from the literature, by supporting the search for relevant articles, the semi-automated extraction of relevant passages, and the transformation of the information from the textual format to a semantic representation.The SASEBio project focuses on the study and development of supporting tools for the semi-automated enrichment of biomedical literature and related documentation, in particular their potential role in the professional literature curation process. Curation is the activity performed by professionals who are paid to read the literature in search of particular items of information and store such information in public databases, which can be accessed later by the biologists.The project deals with entities and relationships that form complex Knowledge graphs, which are gaining relevance to initiate, guide and confirm research in the biosciences.The project leverages on the latest results in Natural Language Processing techniques (combining statistical approaches with human language expertise), in particular information extraction, including biomedical entity and relation detection. While fully unsupervised extraction of information of the literature is currently still unrealistic, text mining tools are already sufficiently reliable to be used to provide hints to the curators, in order to speed up their activities.The results of the project are of academic and industrial relevance for the knowledge management process. The pharmaceutical industry is facing an ever growing complexity of the knowledge space, e.g. genes, proteins and their relationships, as described in the literature. Various research groups maintain legacy systems containing valuable data which cannot be accessed because there is no shared semantic layer. The collaboration between the NITAS/TMS group at Novartis and the OntoGene group at the University of Zurich offers an interesting opportunity to successfully contribute to the solution of the data federation problem. Combining text mining and annotation tools, with the support of a novel web-based text mining application, will accelerate the generation or enrichment of the biomedical data.

Direct link to Lay Summary

Last update: 21.02.2013

Active participation

Self-organised

Number	Title	Start	Funding scheme

SASEBio: Semi-Automated Semantic Enrichment of Biomedical LiteratureThe OntoGene group at the University of Zurich has developed efficienttechniques for text mining in the molecular biology domain. One oftheir core interests in recent years has been the detection of mentionsof protein-protein interactions. Using the IntAct database as a goldstandard, they have developed techniques for the identification ofinformation relevant to the process of curation, such as theexperimental methods used by the authors [1], the organism which arehosts of the experiment and which contribute the interacting proteins[2], the protein themselves [3], and their interactions [4].The effectiveness of their approach has been validated byparticipation to numerous shared evaluations, such as BioCreative II[5], BioNLP event extraction task [6], and BioCreative II.5[forthcoming]. Recently, in collaboration with the NITAS group atNovartis, they have developed an interesting prototype of anenvironment supporting the process of semi-automated semanticenrichment of the literature. The environment allows an expert user toefficiently revise annotations suggested by the system, or to add newannotations where the system missed an entity or an interaction. Thesystem is also capable of reusing the annotations added by the expertin subsequent applications, using a process of incremental learning.The SASEBio project aims at consolidating the existing text miningactivities of the OntoGene group, by further improving their relationextraction techniques, and applying them to new areas, within thecontext of the literature curation process. New types of interactions,such as drug/diseases (of particular interest to their industrialpartner) will be considered, along with incremental improvements totheir existing techniques for protein-protein interaction detection(of potential interest to the IntAct group at EBI). As in the past,their techniques will be subject to community-based evaluation throughparticipation in shared text mining challenges.Additionally, the project offers an opportunity to turn the existingsemi-automated annotation prototype into a fully fledged system whichcan then be employed by the target user groups. Intensivecollaborations with both NITAS and EBI will be sought at all stagesof development, in particular to guarantee a continuous feedback onthe effective usability of the proposed tools.References[1] Thomas Kappeler, Simon Clematide, Kaarel Kaljurand, GeroldSchneider, Fabio Rinaldi. Towards Automatic Detection of ExperimentalMethods from Biomedical Literature. Third International Symposium onSemantic Mining in Biomedicine (SMBM 2008).[2] Thomas Kappeler, Kaarel Kaljurand, Fabio Rinaldi. TX Task:Automatic Detection of Focus Organisms in BiomedicalPublications. BioNLP workshop, NAACL/HLT, Boulder, Colorado 2009.[3] Kaarel Kaljurand, Fabio Rinaldi, Thomas Kappeler, GeroldSchneider. Using existing biomedical resources to detect and groundterms in biomedical literature. Artificial Intelligence in Medicine,Verona, July 2009.[4] Gerold Schneider, Kaarel Kaljurand, Thomas Kappeler, FabioRinaldi. Detecting protein-protein interactions in biomedical textsusing a parser and linguistic resources. CICLING 2009.[5] Fabio Rinaldi, Thomas Kappeler, Kaarel Kaljurand, GeroldSchneider, Manfred Klenner, Simon Clematide, Michael Hess, Jean-Marcvon Allmen, Pierre Parisot, Martin Romacker, Therese Vachon. OntoGenein BioCreative II. Genome Biology, 2008, 9:S13.[6] Kaarel Kaljurand, Gerold Schneider and Fabio Rinaldi. A dependencybased approach to the BioNLP 2009 Shared Task. BioNLP workshop,NAACL/HLT, Boulder, Colorado, 2009