Fig. 8
From: Analysis and correction of compositional bias in sparse sequencing count data
Wrench scales lead to reduced false positive calls. a The average of log2(FOther/FWrench) values obtained over artificial two group splits of homogeneous experimental group data is shown and b the average of log2(COther/CWrench) values across 41 metagenomic experimental groups are shown. Standard error bars are shown. In both plots, positive values for a method imply reduced accuracy relative to Wrench. FOther: total number of diffferentially abundant features found by a competing method (total sum, TMM, CSS or Scran). FWrench: total number of differentially abundant features found by Wrench. COther: total number of features where the covariate term for Wrench normalization factors were found to be significant when competing method is used as offset. CWrench: total number of features where the covariate term for a competing method’s normalization factors were found to be significant, when Wrench is used as covariate