A Methodology to Develop Knowledge Graphs for Indication Expansion: An Exploratory Study

The rapid increase in data sources for computational drug discovery have resulted in the subject of ontology research becoming a key point for not only data integration, but they also have enriched the semantics of biological networks. Drug repurposing using populated ontologies, in other words semantic knowledge graphs, have received much attention in the past years. Firstly, researchers constructed knowledge graphs from biological data sources to predict disease-drug relations. After considering the reliability of the data, these studies are followed by the construction of knowledge graphs from published unstructured data via text mining tools. Only two recent examples are available utilizing PubMed abstracts to build knowledge graphs for drug discovery. However, these studies highlight their limitations due to dependency to the text mining tools and missing information in the abstracts. In light of this previous work, we conducted an exploratory study to develop a novel method for constructing knowledge graphs for indication expansion studies in which the aim is to find an alternative indication for the main target. The implication of the study is that the knowledge graph consists of both biological data sources which have publication references and human curated text mining results from the full texts. Consequently, the prediction results of the methodology includes one or more publication references. This paper presents the methodology together with the application on two selected cases. Moreover, we share the results, lessons learned and future work.