Figure 6.

Predictive Value of IgA Response for IBS-D

(A–F) Machine learning (GBM) evaluation of IBS-D using OTU frequency versus Ig ratio data (see Method Details) using data from Figure 5.

(A) Average prediction values for OTU frequency and Ig-enrichment ratios for Ctrl versus IBS-D, and IBS-D versus IBD (Student’s t test). Note that Ig enrichment is used as IBD patients have IgG⁺IgA⁻ bacteria.

(B) OTUs with the greatest relative influence differ between GBM modeling using OTU frequency versus Ig ratios. Average relative influence values are shown for Ctrl versus IBS-D comparisons with the indicated dataset. The red dots by the OTU name indicate bacteria identified as IgA enriched (Figure 5F).

(C) IgA ratios have a higher area under the curve (AUC) than the OTU frequency for predicting IBS-D versus Ctrl. AUC curves are shown for the indicated comparisons between patient groups using the OTU frequency or Ig ratio data.

(D) Association of IgA frequency with correct disease prediction. GBM prediction values are plotted against IgA/G-bound bacterial percentage.

(E) Correlation of IBS-D prediction with PHQ-15 and HADS anxiety scores. Linear regression analysis of PHQ-15 (subset of STL, n = 12 patients who returned survey) or HADS anxiety (STL + ROC, n = 12 + 22 patients who returned survey) scores versus GBM prediction values from OTU frequency or IgA ratios is shown.

∗p < 0.05, ∗∗p < 0.005, and ∗∗∗∗p < 0.00005. Each experiment on each patient performed once. See also Figures S6 and S7.