Title: Knowledge Graphs for Processing Scientific Data: Challenges and Prospects

Abstract: There is growing interest in the use of Knowledge Graphs (KGs) for the representation, exchange, and reuse of scientific data. While KGs offer the prospect of improving the infrastructure for working with scalable and reusable scholarly data consistent with the FAIR (Findability, Accessibility, Interoperability, and Reusability) principles, the state-of-the-art Data Management Systems (DMSs) for processing large KGs leave somewhat to be desired. In this paper, we studied the performance of some of the major DMSs in the context of querying KGs with the goal of providing a finely-grained, comparative analysis of DMSs representing each of the four major DMS types. We experimented with four well-known scientific KGs, namely, Allie, Cellcycle, DrugBank, and LinkedSPL against Virtuoso, Blazegraph, RDF-3X, and MongoDB as the representative DMSs. Our results suggest that the DMSs display limitations in processing complex queries on the KG datasets. Depending on the query type, the performance differentials can be several orders of magnitude. Also, no single DMS appears to offer consistently superior performance. We present an analysis of the underlying issues and outline two integrated approaches and proposals for resolving the problem.