Skip to main content
Fig. 1 | BMC Genomics

Fig. 1

From: Controlling for conservation in genome-wide DNA methylation studies

Fig. 1

Bias introduced by the Yule-Simpson effect. A comparative analysis of DNA methylation across a functional boundary. A pair of matrices is associated to a functional boundary (one for each side of the boundary). The methylation states are represented by numeric values at all relevant genomic locations (here 0 and 1 are used for simplicity, but any values may be used). For example, row five may represent a genomic locus: CGAGTCAG||CGCGCGCG (|| marking the functional boundary location) where all CG sites are fully methylated. The mean of the row averages (R) is 0.67 for both matrices: in eight of the 12 regions all CGs are completely methylated, and in the remaining four regions all CGs are completely unmethylated. The mean of the column averages (C) is 0.33 for the left matrix and 0.67 for the right matrix. Although there is no difference in the regional methylation average across the matrices, a difference in the column averages is observed due to a different data distribution: regions that are methylated are represented with fewer CGs in the left matrix than they are in the right matrix, but this is not the case for unmethylated regions. Importantly, it is not the lower frequency of sites in the left matrix that causes the discrepancy between C and R but the lower frequency of sites in the methylated regions

Back to article page