CST and taxa regression

We paired microbiome samples from different body sites that were acquired at the same visit for each participant. This generated three pairs of sites. The nasal-rectal sites had the greatest number of matched sample pairs, with 82 participants having 951 pairs of samples, while the nasal-throat sites had the fewest, with 40 participants having 483 sample pairs. The rectal-throat sites had 491 sample pairs. We applied arcsine sqrt-transformation to stabilize the variance of relative abundance and then fit linear mixed effects models to the abundance using the CST of the other two sites as the primary variables of interest. We adjusted as potential confounders the mode of delivery, GA at birth and 14 degree-of-freedom spline for WOL. The subject ID served as a random intercept. Associations with the primary variables of interest were tested for all taxonomic levels and reported on the most specific taxon (or equally specific taxa) within a phylogenetic lineage. We report a Wald test for equality between the abundance in each CST and its grand mean abundance. Associations significant at 5% FDR, calculated per site, are shown as edges in Fig. 5, which itself was generated using R packages GGally version 1.3.2 and network version 1.13.0. An overall test for association between a site and a taxon was derived by conducting an F test of the model that dropped the CST of that site as predictors and the full model. The change in pseudo R² reports the change in variance explained by the fixed effects in the null and full models [92].