Fig. 3: Tumor-normal, clinical, and demographic associations with the microbiome.

a ANCOM-BC differential abundance results with the Holm method for multiple testing correction, and b comparison of Shannon alpha diversity between paired tumor and normal samples using genus-level 16S data (n = 385 tumor-normal pairs) using a two-sided Wilcoxon test. Boxplot centers, upper and lower bounds, and whiskers represent median, upper and lower quartiles, and quartiles ± 1.5 inter-quartile range, respectively. c ANCOM-BC differential abundance results with Holm method multiple testing correction, and d comparison of Shannon alpha diversity between paired tumor and normal samples using genus-level RNA-seq data (n = 525 tumor-normal pairs) using a two-sided Wilcoxon test. Boxplot centers, upper and lower bounds, and whiskers represent median, upper and lower quartiles, and quartiles ± 1.5 inter-quartile range, respectively. e Genus-level alpha diversity and richness in tumors associated via generalized linear models with clinical features, adjusted for study site. RNA-seq (n = 661), 16S (n = 572), and meta-analyzed WGS (n = 704) samples were rarefied to 500, 250, and 100 bacterial reads, respectively. Stage I tumors, adenocarcinoma histology, and European (EUR) ancestry serve as references. Unadjusted p values are shown; all tests are non-significant (FDR > 0.05) after multiple testing correction. Points represent regression coefficient, error bars signify standard error. WGS whole genome sequencing, RNA-seq RNA sequencing, 16S 16S rRNA gene sequencing, AMR American, EAS East Asian.