Enriching Clinical Sample Analysis with Biological Knowledge Graphs:A Preliminary Study

Biological researchers often face challenges in analyzing clinical samples due to the limited amount of samples they can collect. This issue hinders the use of traditional statistical methods; instead, they often rely on their domain knowledge to guide the exploration of the data. To ease the task, we aim to develop a system to support the researchers by integrating data from biological knowledge graphs (KGs), such as Reactome and UniProt, which can drive data exploration through recommendation system techniques. In this article, we present the first step towards such a system by studying whether the data from biological KGs can be represented through embeddings so that proteins of interest are organized and categorized according to their shared functionalities.<br/>We propose Cluster-GAE, a graph autoencoder method inspired by Cluster-GCN. Cluster-GAE combines graph sampling techniques with Graph Neural Networks (GNNs) to learn embedding representations from large-scale biological networks. Our experiments suggest that Cluster-GAE produces embeddings with properties similar<br/>to the embeddings of traditional KG embedding methods without the need to process the whole graph at once.<br/>Moreover, the experiments show that the embeddings produced by Cluster-GAE are of a higher quality than the<br/>embeddings produced by a KG obtained through traditional sampling techniques, in particular Random Walk and<br/>Forest Fire. Finally, through t-SNE visualization and functional enrichment analysis, we showcase the ability of<br/>Cluster-GAE to identify protein clusters that are related to different biological processes, molecular functions,<br/>and cellular components.