Extension of the Interaction Network Ontology for Literature Mining of Gene-gene Interaction Networks from Sentences with Multiple Interaction Keywords

The Interaction Network Ontology (INO) has been demonstrated to be valuable in providing a structured ontological vocabulary for literature mining of gene-gene interactions from biomedical literature. Our analysis of the Learning Logic in Language (LLL) challenge and vaccine datasets showed that many interactions are signaled with 2 or more interaction keywords used in combination. In this paper, we extend the INO by adding combinatory patterns of two or more literature mining keywords to related INO interaction classes. An INO-based literature mining pipeline was further developed based on SPARQL queries and SciMiner, an in-house literature mining program. The majority of the gene interaction sentences from the LLL and vaccine datasets were found to use multiple keywords to represent interaction types. A comprehensive analysis of the LLL dataset identified 27 gene regulation interaction types each associated with multiple keywords. Special patterns were discovered from the hierarchical structure of these 27 INO types.