Drug repurposing through computational biomedicine is a popular and actively researched field. It involves the use of Machine Learning, Natural language processing, network analysis, etc. to find previously undiscovered links between pre-existing drugs and new diseases. One such model is the KGML-xDTD (Ma et al., 2023) that utilizes a biomedicine knowledge graph (the RTX-KG2 in this case (Wood et al. 2022)) to output drug-disease pairs and corresponding mechanism of actions. There have been multiple versions of the knowledge graph, and thus multiple predictions made by the ML model. In this paper, I analyze the variation in knowledge graphs over time and the results of the KGML-xDTD model on each version of the knowledge graph. We will see that similarity between graphs seems to correlate with the duration between when they were made. Moreover, I will summarize attempts to implement ML ensemble techniques, which did not result in meaningful results. Finally, this paper will present a possible method of network analysis with the help of community analysis of the knowledge graph. Community finding seems to show promise in finding similar nodes and notable paths in the graph.
