Figure 3.

Normalized expression levels in a random set of genes with both UM and gbM epialleles. To make this figure, a random set of 100 genes was selected and only genes with both UM and gbM epialleles were retained. Genes with median read number under 10 were also excluded, leaving 57 genes. For each gene, the box plots indicate normalized expression levels for gbM and UM epialleles, with the lines representing the medians and the width of the boxplot the 1st and 3rd quartile. The figure shows that accessions that have a gbM epiallele tend to have a higher median normalized expression compared to accessions that have a UM epiallele, because 34 out of 57 genes in the represented random gene set have higher median expression levels for gbM epialleles. Although the differences are small and some genes show the opposite pattern, the overall effect is significant in a linear model across genes (see text for details). For each accession, normalized expression level was computed as follows: (accession normalized read number—median gene normalized read number)/median gene normalized read number.