Evaluating trait-based sets for taxonomic enrichment analysis applied to human microbiome data sets

BackgroundSet-based pathway analysis is a powerful tool that allows researchers to summarize complex genomic variables in the form of biologically interpretable sets. Since the microbiome is characterized by a high degree of inter-individual variability in taxonomic compositions, applying enrichment methods using functionally driven taxon sets can increase both the reproducibility and interpretability of microbiome association studies. However, there is still an open question of which knowledge base to utilize for set construction. Here, we evaluate microbial trait databases, which aggregate experimentally determined microbial phenotypes, as a potential avenue for meaningful construction of taxon sets. MethodUsing publicly available microbiome sequencing data sets (both 16S rRNA gene metabarcoding and whole-genome metagenomics), we assessed these trait-based sets on two criteria: first, do they cover the diversity of microbes obtained from a typical data set, and second, do they confer additional predictive power on disease prediction tasks when assessed against measured pathway abundances and PICRUSt2 prediction. ResultsTrait annotations are well annotated to a small number but most abundant taxa within the community, concordant with the concept of the core-peripheral microbiome. This pattern is consistent across all categories of traits and body-sites for whole genome sequencing data, but much more heterogenous and inconsistent in 16S rRNA metabarcoding data due to difficulties in assigning species-level traits to genus. However, trait-set features are well predictive of disease outcomes compared against predicted and measured pathway abundances. Most important trait-set features are more interpreable and reveal interesting insights on the relationship between microbiome, its function, and health outcomes.