Biomedical Event Extraction with Hierarchical Knowledge Graphs

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1277–1285 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 1277 Biomedical Event Extraction with Hierarchical Knowledge Graphs Kung-Hsiang Huang 1 Mu Yang1 Nanyun Peng1,2 1 Information Sciences Institute, University of Southern California 2 Computer Science Department, University of California, Los Angeles {kunghsia, yangmu}@usc.edu violetpeng@cs.ucla.edu Abstract Biomedical event extraction is critical in understanding biomolecular interactions de- scribed in scientific corpus. One of the main challenges is to identify nested structured events that are associated with non-indicative trigger words. We propose to incorporate do- main knowledge from Unified Medical Lan- guage System (UMLS) to a pre-trained lan- guage model via a hierarchical graph repre- sentation encoded by a proposed Graph Edge- conditioned Attention Networks (GEANet). To better recognize the trigger words, each sentence is first grounded to a sentence graph based on a jointly modeled hierarchical knowl- edge graph from UMLS. The grounded graphs are then propagated by GEANet, a novel graph neural networks for enhanced capabilities in inferring complex events. On BioNLP 2011 GENIA Event Extraction task, our approach achieved 1.41% F1 and 3.19% F1 improve- ments on all events and complex events, re- spectively. Ablation studies confirm the impor- tance of GEANet and hierarchical KG. 1 Introduction Biomedical event extraction is a task that identifies a set of actions among proteins or genes that are as- sociated with biological processes from natural lan- guage texts (Kim et al., 2009, 2011). Development of biomedical event extraction tools enables many downstream applications, such as domain-specific text mining (Ananiadou et al., 2015; Spangher et al., 2020), semantic search engines (Miyao et al., 2006) and automatic population and enrichment of database (Hirschman et al., 2012). A typical event extraction system 1) finds trig- gers that most clearly demonstrate the presence of events, 2) recognizes the protein participants (ar- guments), and 3) associates the arguments with the corresponding event triggers. For instance, the Is a Component of� possibly included Phosphorylation Smad BMP-6 Induce Mechanism of� action qualifier Functional� concept Protein Biological function Physiologic� function BMP-6 induces phosphorylation of Smad 1/5/8 Trigger labels Argument labels Sentence Theme Theme Cause Positive Regulation Protein Phosphorylation Protein Can be qualified by Action of Positive regulation of biological process Cause of Allowed qualifier Figure 1: An example of a UMLS-based hierarchi- cal KG assisting event extraction. Circles represent concept nodes and triangles represent semantic nodes. Nodes associated with the tokens in the example sen- tence are boldfaced. Bidirectional edges imply hier- archical relation between concept and semantic nodes. The word “induces” is a trigger of a Positive regulation event, whose trigger role and corresponding argument role cannot be easily determined with only textual in- put. The KG provides clues for identifying this trigger and its corresponding arguments given the red and blue double line reasoning paths connecting nodes BMP-6, Induce, Phosphorylation, and Positive regulation of biological process. We can infer that: 1) “induces” is an action of biological function, 2) a biological func- tion can be quantified by …

Content truncated to 500 words