Fig. 4.

Differentially abundant molecular clusters and microbe-metabolite co-occurrences in obesity-inflammation-depressive and inflammation status. a Sample plot showing log-ratio of differential molecular features relative to cluster 1 (see left panel). The corresponding right panels represent a scatterplot of samples showing log-ratio of differential features versus inflammation status. Individual samples are colored by health status. Statistical significance of the log-ratios was evaluated by pairwise comparisons using Wilcoxon rank sum test. A linear regression model was employed for regressing log-ratios against BARIC values. b Visualization of microbe-metabolite co-occurrences. Arrows represent microbes and dots represent metabolites. The x and y axes represent principal components of the microbe-metabolite conditional probabilities as determined by the neural network. Distances between arrow tips quantify co-occurrence strengths between microbes, while directionality of the arrows indicates which microbes and metabolites have a high probability of co-occurring. Only known microbiota-derived molecules are labeled. Microbial abundances are estimated using differential abundance analysis via multinomial regression