Graph Representation Learning for IBD Diagnosis Based on Microbiome Metaomic Data

Christopher Irwin, Flavio Mignone, Stefania Montani, Luigi Portinale

Abstract: Inflammatory Bowel Disease (IBD) is a chronic relapsing inflammatory disease strictly related to the gut microbiome, a critical component of human health. The analysis of gut microbiome presents substantial challenges due to the high dimensionality and sparsity of its metaomic data. Traditional approaches often struggle to uncover the intricate relationships within microbial communities. In this work, we introduce a Graph Representation Learning (GRL)-based framework designed to classify patients based on the presence or absence of IBD conditions. Unlike conventional methods that rely primarily on taxa abundance, our approach directly incorporates taxonomic relationships, enabling the construction of a generalized encoder for microbial taxa networks. The learned representations effectively capture the structure and dynamics of individual gut microbiomes, which are subsequently utilized for phenotype prediction (i.e., presence or absence of IBD). By bridging microbial taxa relationships with patient classification, our method demonstrates the potential to advance microbiome-based diagnostics and deepen our understanding of the role microbial communities play in disease states.