Fig. 4

Identification of gut microbiota compositional features and biomarkers of host-microbiome interactions that predict satiety and surrogate endpoint responses by machine learning. (left) AUC-ROC curves show the performance accuracy of random forest classifiers trained to predict high-vs-low responders for: A and B perceived satiety after a meal with AX using the relative abundance of bacterial taxa activated during ex vivo incubation with AX; C HOMA-IR and D fecal calprotectin for AX and MCC, respectively, using fecal bile acid shifts; and E TNF-α for MCC using baseline intakes of calorie-adjusted macronutrients. (center) Horizontal bars represent Spearman’s correlation coefficients between endpoints and A and B metabolically active ASVs, C and D fecal bile acids, or E macronutrients shown to be important for predicting responses. Mean importance values were determined by random forest, which identifies factors that contribute the most to the model. (right) Scatter plots show the association between endpoints and the most discriminative microbiota-related factors that correlate with AX-induced A and B satiety after a meal and C HOMA-IR attenuation, and D MCC-induced fecal calprotectin attenuation. Vertical bar graphs show the most discriminative microbiota-related factors grouped by high- and low-responders. High-responders (black) and low-responders (gray) were defined according to the study cohort median. Statistical significance was set at p < 0.05 and FDR adjusted q values < 0.05. ∆, absolute change from baseline to week 6; %∆, percent change from baseline to week 6; ³√, cube root transformed before analysis; All ASVs, amplicon sequence variants with average relative abundances ≥ 0.15%; AX, arabinoxylan; AUC-ROC, area under the receiver operating characteristic curve; BL, baseline; Diff. Abundant ASVs, differentially abundant amplicon sequence variants among the bacterial consortia recovered by fluorescence-activated cell sorting; GDCA, glycodeoxycholic acid; HOMA-IR, homeostatic model assessment of insulin resistance; ILCA, isolithocholic acid; LCA, lithocholic acid; MCC, microcrystalline cellulose; TLCA, taurolithocholic acid; TNF-α, tumor necrosis factor-α