A Phenotype-Embedded Mapper Framework Links Microbiome-Metabolome Interaction Modules to Colorectal Cancer.

Integrative analysis of the gut microbiome and metabolome can help characterize colorectal cancer (CRC)-associated molecular changes that are difficult to resolve from either omics layer alone. However, microbiome-metabolome data are high-dimensional, heterogeneous, and often contain nonlinear or locally confined associations that may be obscured by global linear models. Here, we propose a phenotype-guided topological framework that extends the Mapper algorithm for local interpretation of paired microbiome and metabolome profiles. Disease-associated variation from each omics block was summarized by partial least-squares regression and used to construct a two-dimensional filter space for Mapper graph construction. We further developed an Extended Spatial Analysis of Functional Enrichment strategy (eSAFE) to evaluate the spatial enrichment of phenotypes, individual features, and feature-pair associations on the resulting graph. Applied to paired fecal metagenomic and metabolomic profiles from a CRC cohort, the framework organized samples into phenotype-aligned neighborhoods and identified localized microbial, metabolic, and cross-omics association patterns linked to CRC. Coenrichment analysis further prioritized disease-associated features and interaction modules that were partly distinct from those obtained by univariate differential analysis or supervised sparse multiblock integration. One disease-localized microbiome-metabolome module showed moderate CRC discrimination in internal cross-validation and was enriched for metabolites involved in butanoate and amino acid-related pathways. These results suggest that phenotype-guided topological analysis can provide a complementary, interpretable view of localized multiomics organization in CRC-associated gut ecosystems.