Fig. 4
Use of genera to predict eventual response to treatment in pretreatment samples. a Our classifier classifies response status significantly better than random guess with AUC = 0.75 and overall accuracy of 76.5 % for predicting treatment response/non-response. b Box plots of the log10 relative abundance plus pseudocount (1E-05) of the 15 genera with highest importance scores in random forest analysis in responders and non-responders. The asterisks denote taxa also identified as significant in our generalized estimating equations analysis. (See also Additional file 2: Figures S4 and S6, Additional file 2: Tables S7 and S9.)