Using syntactic structure to extract prominent gene regulatory network from the literature

Abstract

The curation of a biomedical database should be conducted automatically as it is too big and growing too fast to take care of manually. Due to the huge size of the data contents and the variety of data type, natural language processing (NLP) techniques are utilized to convert an article to a relational database format without any manual process. However, current biomedical NLP techniques do not consider the syntactic tone as a separate metric to represent the author’s judgment on the prominence of a statement. As the biomedical literature target expert readers, authors often deliver information via a subtle syntactic structure. For example, when an author writes a sentence in present tense, the author is understood to deliver the message that it is an accepted fact. In this work, we analyze the scientific reports for gene regulatory network (GRN), for which manually curated da-tabase for prominent GRN is available. Through the analysis, we verified that the syntax structure is key to distinguish whether or not a report is stating prominent knowledge. We identify the key syntactic features for such distinction, and present a continuous representation of the prominence degree of a knowledge statement into a real number, solely calculated by the syntactic structure of the statement. In addition, we argue for the importance of such research for an effective knowledge discovery platform.