Skip to main content


Fig. 10 | BMC Genomics

Fig. 10

From: Analysis and correction of compositional bias in sparse sequencing count data

Fig. 10

Wrench retains potential biological information, and indicates importance of compositional correction in general practice. We plot some statistical summaries and the compositional scale factors reconstructed by a few techniques for various Human Microbiome Project samples, sequenced at the Baylor College of Medicine. a On the top-left, we plot the logged median of the positive ratios of group-averaged proportions to that of Throat chosen as the reference group. Stool samples show considerable deviation from the rest of the samples despite having comparable fraction of features detected and sample depths to other body sites. Notice the log scale. b The similarity in the reconstructed scales across techniques (second row) for closely related body sites are striking; although minor variations in the relative placements were observed across centers potentially due to technical sources of variation, the overall behavior of highly significant differences in the scales of behind-ear and stool samples were similar across sequencing centers (Additional file 1: Figure S10) and normalization methods. Corresponding CSS scales in Additional file 1: Figure S11. These techniques predict a roughly 4X-8X (ratio of medians)inflation in the Log2-fold changes when comparing abundances across these two body sites. c Wrench and scran compositional scale factors across the plant-based diet (BK) and Western diet (Western) mice gut microbiome samples. d Compositional scale factors for healthy (Control) and diarrhea afflicted (Case) children. Slight differences in the compositional scales are predicted in the diet comparisons with t-test p-values < 1e-3 for all methods except TMM, but not as much in the diarrheal samples

Back to article page