Fig. 1

Data processing pipeline. A Displayed are the processing steps applied for data preprocessing (cleaning, prevalence filtering, CLR transformation and adjustment for nuisance variables) and analysis. B Graphical display of the main steps for network construction and motif analysis. From left to right: (1-left) Computation of the joint Bayesian correlation matrix. Because for undirected graphs, the correlation matrix is symmetrical, and only their upper part is considered in further steps. Colour codes the strength of evidence for a certain connection with white to red preferring H1 and white to blue preferring H0. Sections associated with psychopathology (PP, depression and fatigue), taxonomical (T, genera) and metabolic (M, KEGG modules) abundances are separated by thin black lines. (2-middle) To binarise this matrix, a threshold of Log₁₀BF10 ≥ 0.5 was applied. For the remaining matrix elements or node connections, H1 is at least 3 times more likely than H0. The resulting binary adjacency matrix was used to construct the association network. (3-right) Exemplary representation of one triangular motif of interest, that is composed of interconnected nodes of all three modalities