Molecular Interaction Evidence Fragment Corpus

This file set showcases the first presentation of an annotated text corpus of 'evidence fragments' from biomedical papers. These are short passages of text from the results sections of papers that describe the goals, methods, results and interpretations of experiments that are reported as figures in biomedical articles. These figures typically provide the evidence and justification for any underlying scientific explanation or argument being explored in any given paper. Moreover, these fragments describe the specific findings of experimental studies and therefore comprise the 'raw material' of investigative scientific work. It is for that reason that we prioritize them within this corpus. <br><br>The file set should be considered as consisting of two parts: (1) a single archive of annotated text fragments forming the 'Molecular Interaction Evidence Fragment Corpus' (intact_evidence_fragment_corpus_07-25-17.tar.gz); and (2) other files provided for provenance and reproducibility. <br><br><b>The 'Molecular Interaction Evidence Fragment Corpus'<br></b><br>This gzipped tar archive forms the initial preparation of distant learning data corpus where we develop a method to link text to data. Here we use references to figure labels ('Fig 2a', etc) to identify (A) excerpts from the results sections of papers describing the goals, methods, results and interpretations of experiments showcased in those figures. We call these passages 'evidence fragments' and are contained in *.ttl files that use an OWL-enabled version of the National Library of Medicine's BioC standoff annotation format (http://bioc.sourceforge.org/) with additional relations from the SPAR ontologies (http://sparontologies.net/) to link these fragments to figure references. (B) We also use these same figure references to anchor manually-curated INTACT data records where noted (http://www.ebi.ac.uk/intact/). We restricted our analysis to papers from the PMC open access subset. <br><b>Provenance Data Files.<br></b><b><br></b>We provide a set of files for the purposes of provenance and reproducibility. This includes the following subsets of papers:<br>* <b>nxml</b> - INTACT open source articles processed in this paper downloaded from PMC.* <b>train+test</b> - Training and testing data for SciDT in TSV form. The articles forming this small corpus are derived from open source papers indexed in the Pathway Logic database developed by SRI International. * <b>scidt_bioc</b> - data fully processed by the SciDT pipeline and saved in BioC format* <b>gannt</b> - 'gantt' charts for each paper that show the distribution of evidence fragments over clauses in the results sections of papers. * <b>intact_evidence_fragment_corpus_10-17-17.tar.gz </b>- a tar archive of <i>*.ttl</i> files for both biopax and bioc sections of the data. We also include <i>*.jsonld</i> formatted files for the bioc annotations. <br>http://purl.org/ske/ro/semsci17 points to a research object (http://www.researchobject.org) that links to the software used to generate this data. <br>