Fig. 4: Association between microbiota development and oral rotavirus vaccine seroconversion.
A Longitudinal analysis of alpha diversity. Shannon index was calculated at genus level. Cross-sectional comparisons were performed using logistic regression. Longitudinal comparisons were performed using mixed-effects models including week of life as a covariate and study ID as a random effect. See Fig. 1 legend for box plot parameters. B Proportion of variation in microbiota composition associated with seroconversion. R2 and statistical significance were determined by PERMANOVA using genus-level unweighted Bray–Curtis distances. C Cross-validation accuracy of Random Forests for prediction of seroconversion. Models were fitted at genus and RSV level. Median out-of-bag accuracy (proportion correctly assigned) and interquartile range across 20 iterations of 5-fold cross-validation are displayed. Each iteration included an equal number of responders and non-responders (50 per group where possible, or else the number in the minority group if this was <50). For each cross-sectional comparison, taxa were classified as discriminant if they had an FDR-adjusted p value of <0.05 based on either two-sided Fisher’s exact test (differences in prevalence) or Aldex2 with two-sided Wilcoxon rank-sum test (differences in abundance). CV, cross-validation; IND, India; MLW, Malawi; neo+, infected with rotavirus neonatally (defined by detection of rotavirus shedding in week of life 1 or baseline seropositivity); neo−, uninfected with rotavirus neonatally; nf, not fitted as <10 responders; ns, not significant; RF, Random Forests; RSV, ribosomal sequence variant; *p < 0.05; **p < 0.005; ***p < 0.0005.
