Extracting Semantics of Predicates From Millions of Bio-Medical Abstracts for Inferencing New Biological Key Events and Relationships

Adverse outcome pathways (AOP) structure toxicological knowledge as sequential, directed chains of key events (KE) that culminate in adverse outcomes (AOs). AOP development is a laborious process that involves extensive knowledge mining and could be improved via use of machine / deep learning. In this paper, we present an artificial intelligence system that can accelerate putative AOP development process by inferencing new AOP modules based on the knowledge learned from 16-million pre-parsed PubMed abstracts. Each AOP modules is represented as a triplet that consists of antecedent and consequent biological entities connected by a relation. Users can also investigate specific types of antecedent, consequent, and relations by specifying macro/microtemplates using the MeSH semantic type hierarchy. We also provide visualizations to illustrate the hidden semantics that our system can extract from input triplets.