Link Prediction for Biomedical Network

Network datasets are seen ubiquity in many fields, such as protein interactions, paper citation, and social networks. While some networks are well-defined, many others are not. For example, the interactions of proteins in cancer pathways are still studied by system biologists and medical researchers. Therefore, one of the primary analytic tasks to perform on these networks is link prediction, where we desire to reveal some unknown relationships with certain levels of confidence. In this paper, we carry out some experiments on network datasets in the biomedical domain using state-of-the-art Graph Neural Networks. The results show that entity’s values facilitate graph-based models to perform well on uncovering latent relationships in biomedical research and potentially be extended on other application domains.