Relative abundances of specific genera at time of ALL diagnosis deviate from healthy siblings and determine risk of neutropenia after induction treatment. (A) Bacterial genera with a prevalence >25% and a significant difference (P < .05) in relative abundance between patient samples on day 1 and healthy siblings. (B) sPLS regression for the relative abundance of bacterial genera (prevalence >25%) on day 1 and prolonged neutropenia (n = 12/32). Bacteria are sorted from top to bottom based on their sPLS model loading (lower to higher). The brown and red bars indicate negative and positive loadings, respectively. The genera marked with darker brown and ∗∗ are shown in panel C. (C) The median (range) of AUC from 5-time repeated 10-fold cross-validation of the sPLS model. Box plots represent class predictions for the median cross-validation model. Boxes show the median level with the 25th and 75th percentiles and whiskers represent the range. (D) Relative abundance on day 1 of the 10 genera that turned out with significant difference in relative abundance between the groups (P < .05). Boxes show the median level with the 25th and 75th percentiles; whiskers and outliers represent the 5th and 95th percentiles and range, respectively. P values (A, D) correspond to Wilcoxon rank-sum tests without correction for multiple tests. ∗ represents FDR-corrected P values. CV, cross-validation.