GePI: Retrieval of fully automated recognition and extraction of gene and protein interaction mentions from unstructured literature

Abstract

AbstractMotivationKnowledge about interactions between genes and proteins is vital for bio-molecular research. A large part of this knowledge is published in written text and not accessible in a structured way. To remedy this situation, several repositories of automatically extracted interaction facts were proposed over the years. However, existing solutions lack key features such as permanently updated data resources, easy accessibility and structured result generation ready to be used for downstream analyses.ResultsWe propose GePI, a database portal for fully automated extraction and presentation of molecular interaction facts from scientific literature. GePI offers batch queries, immediate inspection of textual evidence and full text filters. To this end, GePI leverages two gene recognition and normalization approaches as well as optimized runtime for molecular event extraction. The resulting natural language processing pipeline is applied to the full set of publicly available documents from PubMed and the PubMed Central open access subset accounting for more than 33M abstracts and 4.2M complete articles as of 2022. To accommodate the rapid growth of the scientific literature, the fact database is automatically updated several times per week. In summary, our web application GePI allows for the first time a free and easy-to-use investigation of gene and protein interaction information as soon as they are published with unique query possibilities.Availability and ImplementationThe GePI web interface is available at http://gepi.coling.uni-jena.de.Contacterik.faessler@uni-jena.de