Building the Scientific Knowledge Mine (SciKnowMine): a community-driven framework for text mining tools in direct service to biocuration

Although there exist many high-performing text-mining tools to address literature biocuration (populating biomedical databases from the published literature), the challenge of delivering effective computational support for curation of large-scale biomedical databases is still unsolved. We describe a community-driven solution (the SciKnowMine Project) implemented using the Unstructured Information Management Architecture (UIMA) framework. This system's design is intended to provide knowledge engineering enhancement of pre-existing biocuration systems by providing a large-scale text-processing pipeline bringing together multiple Natural Language Processing (NLP) toolsets for use within well-defined biocuration tasks. By working closely with biocurators at the Mouse Genome Informatics (MGI) group at The Jackson Laboratory in the context of their everyday work, we break down the biocuration workflow into components and isolate specific targeted elements to provide maximum impact. We envisage a system for classifying documents based on a series of increasingly specific classifiers, starting with very simple surface-level decision criteria and gradually introducing more sophisticated techniques. This classification pipeline will be applied to the task of identifying papers of interest to mouse genetics (primary MGI document triage), thus facilitating the input of documents into the MGI curation pipeline. We also describe other biocuration challenges (gene normalization) and how our NLP-framework based approach could be applied to them. 1 The SciKnowMine project is funded by NSF grant #0849977 and supported by U24 RR025736-01, NIGMS: RO1-GM083871, NLM: 2R01LM009254, NLM:2R01LM008111, NLM:1R01LM010120-01, NHGRI:5P41HG000330 2 http://www.informatics.jax.org