Skip to main content


Fig. 2 | BMC Genomics

Fig. 2

From: aFold – using polynomial uncertainty modelling for differential gene expression estimation from RNA sequencing data

Fig. 2

Fold change shrinkage of the aFold and DESeq2 methods. Results are based on the SEQC data set. Fold change is studied for two test comparisons, generated by randomly combining samples from the SEQC data set (four UHR samples). For the first test comparison in (a), samples from identical conditions are combined (all UHR), resulting in the absence of true DE (labeled “Without DE”; the left and middle panels). The second test comparison in (b) additionally includes pseudo reads count from UHR scaled according to fold change between UHR and BHR, yielding a data set with 40% true DE (labeled “With DE”; right panels). The results are only shown for non-DE genes, in order to enhance comparability between results for data sets with and without DE. This comparison thus allows us to assess the effect of DE genes on normalization efficiency for the non-DE genes. Results for DESeq2 and aFold are based on geometric and qtotal normalization, respectively. They suggest that the presence of a large proportion of DE genes reduces the efficiency of data shrinkage by the approach implemented in DESeq2 but not aFold

Back to article page