Co-occurrence analysis

Bacterial taxonomy was collapsed at the family level, Spearman correlation was calculated between each pair of families using SciPy [83], cluster maps were generated using seaborn (S3 Fig), and ethnic associations were drawn from S5 Table. Correlations were masked where Bonferroni corrected Spearman p-values were >0.05, and clusters were identified as the most prominent (strongest correlations) and abundance enriched. Enrichment of ethnic association was evaluated by measuring the Mann–Whitney U of cluster families’ ethnic associations (p-values, S5 Table) compared to the ethnic associations of noncluster taxa. Cluster-associated families were identified as having at least three significant correlations with families within the cluster.