Enterotyping of Gut Microbiome Composition

Samples were rarefied to 10,000 randomly selected reads, with exclusion of samples with <10 000 reads. Enterotyping (or community typing) was performed based on the Dirichlet-multinomial Model (DMM) approach in R (dmn function) as previously described^{37 (link)} on a combined genus-level abundance matrix containing the study samples as well as 1106 Flemish Gut Flora Project (FGFP)²⁰ samples to increase accuracy. The DMM method applies probabilistic models to bin samples into (a non-predetermined number of) clusters based on their similarity in microbiota composition. The Bayesian Information Criterion (BIC) was used to determine the optimal number of clusters (k = 4; minimum BIC = 429,086.9), and the probability for enterotype assignment was calculated for all samples. Mean probability ± standard deviation for community-type assignment was 0.97 ± 0.089. The four clusters were named after their enterotype-discriminating predominant taxa,^{13 (link)} being Ruminococcaceae (Rum; 29% of samples), Prevotella (Prev; 17% of samples), and Bacteroides 1 and 2 (Bact1 and Bact2; 38% and 16% of samples, respectively).