Fig. 4

Ability of infant or maternal microbiome profiles to predict infant microbiome membership. Both infant and maternal a species-level relative abundance profiles and b functional profiles (ECs) were used as predictors in random forests to infer (1) the presence/absence of other individual species by sequencing, (2) subject-varying S. aureus variables, and (3) subject-fixed S. aureus culture phenotypes in infants. When predicting properties of a given species (presence/absence or any S. aureus-derived property), we a removed the species from the abundance table or b removed any ECs contributed to by the species and renormalized all samples to 100% prior to model fitting to avoid circularity. The infant microbiome exhibited reasonably strong within-ecosystem cohesion (i.e., predictability) but essentially none from the maternal to infant microbiome; conversely, only certain S. aureus carriage phenotypes were well predicted by this model