Interpreting graph neural networks reveals microbial association patterns in inflammatory bowel diseases
Traditional machine learning models trained on gut microbiome abundance profiles can distinguish individuals with inflammatory bowel disease (IBD) from non-IBD individuals. However, their interpretation typically considers taxa in isolation, making it difficult to understand how associations between taxa contribute to predictions. Here, we inferred individual-specific microbial association…
Graph neural networks offer insights into microbial associations in inflammatory bowel diseases (IBD), a notable improvement over traditional machine learning models. While abundance-based methods, such as Random Forest, perform better in prediction accuracy, they fall short in explaining the biological mechanisms behind IBD. By analyzing longitudinal metagenomics data from 106 individuals, researchers inferred individual-specific microbial association networks and trained GNN models for comparison.
The GNN models revealed that a densely connected module of co-abundant Bacteroidales species played a critical role in IBD predictions. Surprisingly, these species were not differentially abundant between conditions, unlike the taxa prioritized by RF. In an independent cohort, this module carried most of the predictive signal, and its removal led to performance comparable to chance.
This finding demonstrates that coordinated association structure, rather than mere abundance shifts, is a key feature of the IBD gut microbiome. More importantly, interpreting GNNs can uncover such patterns that complement abundance-based analyses, providing a more comprehensive understanding of IBD pathogenesis.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
