A Block-Scaled Early Fusion Framework for Multi-Omics Integration Reveals Microbiome-Immune Bridge Features in Inflammatory Bowel Disease

Background: Inflammatory bowel disease (IBD) biomarkers demonstrate substantial inter-individual variability, with 20% of patients showing no elevation despite active disease. Multi-omics integration offers promise for molecular stratification, but existing approaches suffer from modality dominance and high false discovery rates. Methods: We applied block-scaled early fusion principal component analysis to the Integrative Human Microbiome Project dataset, integrating serologic data with metagenomics-derived microbial taxonomic profiles and metagenomics-derived functional enzymatic capacity from 7 controls and 11 IBD cases at temporally matched timepoints. Three network inference strategies (cosine similarity, k-nearest neighbor Louvain, and shared-signal correlation) were used to identify cross-modal bridge features. Results: Block scaling prevented taxonomic modality dominance despite 135 taxa versus 8–11 serologic features. Canonical IBD serology markers (ASCA, ANCA, OmpC, CBir1) emerged as stable immune anchors across health and disease, but microbial partners diverged by disease status. In controls, ASCA bridged to Flavonifractor and oxidative phosphorylation enzymes; whereas in IBD cases, to Peptoniphilus and peptidoglycan biosynthesis (MurB). Serologic features dominated the leading principal component in both groups (77.4% controls, 71.0% cases), while microbial/enzymatic features contributed more to case heterogeneity. Cross-method bridge agreement ranged 0.689–0.893 for nodes spanning ≥2 modalities. Conclusions: This framework mitigates modality dominance and identifies reproducible cross-modal bridges linking microbiome composition, metabolic activity, and immune state to disease status. Temporal matching was used to reduce potential bias arising from asynchronous host and microbial measurements in a biologically dynamic system. This approach provides a hypothesis-generating framework for multimodal biomarker discovery in IBD and warrants validation in larger, prospectively collected cohorts.