Fig. 1

Most diseases show microbiome alterations, and consistent disease-associated shifts differ in their extent and direction. a Left: Total sample size for each study included in these analyses. Additional information about each data set can be found in Table 1. Studies on the y-axis are grouped by disease and ordered by decreasing sample size (top to bottom). Right: Area under the ROC curve (AUC) for genus-level random forest classifiers. X-axis starts at 0.5, the expected value for a classifier which assigns labels randomly, and AUCs less than 0.5 are not shown. ROC curves for all data sets are in Supplementary Fig. 1. Note that Youngster et al.¹⁸ had only four distinct control patients was excluded from the random forest analysis. b Left: Number of genera with q < 0.05 (Kruskal–Wallis (KW) test, Benjamini–Hochberg FDR correction) for each data set. If a study has no significant associations, no point is shown. Right: Direction of the microbiome shift, i.e., the percent of total associated genera which were enriched in diseased patients. In data sets on the leftmost blue line, 100% of associated (q < 0.05, FDR KW test) genera are health-associated (i.e., depleted in patients relative to controls). In data sets on the rightmost red line, 100% of associated (q < 0.05, FDR KW test) genera are disease-associated (i.e., enriched in patients relative to controls). Supplementary Figs. 14 and 15 show q values and effects for each genus in each study