Identifying genetic interaction evidence passages in biomedical literature

In this work, we report our contributions to the BioC Track of BioCreative V for the task of identifying genetic interaction evidence passages. Text describing genetic interactions is difficult to identify due to no simple definition for these interactions and lack of training data. We prepared two manually annotated datasets containing 1793 PubMed abstract and 1000 full text sentences, respectively. We also built two classification systems to identify genetic interaction evidence, one based on word and context features, and one based on query features used for genetic evidence information retrieval. Both models gave satisfactory results on our manually annotated datasets and we produced four different runs, which were submitted for inclusion in the complete BioC Track system. Identification of genetic interactions in biomedical text is a challenging problem with much work still needing to be done.