Fig. 1

C3NA framework illustrated for two conditions comparison examples. For every phenotype/diagnosis, a condition-specific Phyloseq will be used as input to generate the condition-specific stacked-taxa count matrix by combining Phylum, Class, Order, Family, Genus, and Species-level raw count matrix. Then, the matrix undergoes SparCC correlation calculation with 1000 bootstraps followed by the topological overlap matrix (TOM) calculation under the “signed” network setting. Next, the dissimilarity TOM matrix (1-TOM) is used for hierarchical clustering with a range of minimal taxa per module (3–40) to extract a range of clustering patterns. A selected range of patterns is used to generate a consensus matrix, in which the intra-modular connections are the key taxa–taxa correlations we focus on in the subsequent network analysis. When comparing two conditions, module preservation analyses and other statistical methods are performed