SemRep exploits UMLS knowledge resources to identify semantic predications and is made available to the research community as a stand-alone system. Current SemRep research and development involves enhancing linguistic processing for increased accuracy as well as migrating its codebase to Java for more efficient development. SemRep additionally supports two other resources that we provide to the research community. SemMedDB is a PubMed-scale repository of semantic predications extracted from biomedical abstracts (more than 94 million predications as of December 31, 2017). Semantic MEDLINE is a web application that combines semantic predications with document retrieval, abstractive summarization, and visualization techniques and allows users to navigate the biomedical literature using concepts and relations between them. These resources have been widely used as the basis for practical tasks, such as identification of drug interactions and adverse effects, hypothesis generation, and literature-based discovery. Current work on SemMedDB focuses on providing more contextual information about the semantic predications in SemMedDB, including their rhetorical category (IMRAD), factuality status, and evidence level. We are also exploring different distribution options for SemMedDB, such as graph databases and RDF triplestores, to provide new aggregation and reasoning capabilities. Recently, we have also been using information extraction methods more broadly to assess scientific rigor and translatability of biomedical research, in collaboration with extramural researchers. A recent study focused on automatic recognition of limitations in clinical research literature. A follow-up study aims to assess adherence of clinical trial publications to CONSORT reporting guidelines for increased reproducibility. Other work in this area includes extraction of study characteristics (e.g., species, models, interventions, outcomes) from PubMed abstracts to allow a large-scale unbiased analysis of variables relevant to translatability of pre-clinical animal studies, as well as fine-grained citation analysis for enhanced bibliometrics.
