Fig. 3

mQTL mapping results in the baboon RRBS data. a IMAGE identified more mQTL than the other five methods across a range of empirical FDR thresholds. b IMAGE identifies more consistent associations than the other methods in the subset analysis. Here, we randomly split individuals into two approximately equal-sized subsets and analyzed the two subsets separately using each method. We then counted the number of overlapping mQTL identified in both subsets. The overlap ratio (y-axis) is plotted against the percentage of top mQTL ranked by statistical evidence for a SNP-CpG methylation association in each method (x-axis). c Upper panel: log₂ odds ratio of detecting associated SNP-CpG pairs, together with the 95% CI, is computed for CpG sites residing in different annotated genomic regions. CpG sites with IMAGE-identified mQTL are enriched in open sea regions (p value = 0.0106) and depleted in CpG islands (p value = 1.056 × 10⁻⁹). Bottom panel: all analyzed CpG sites were annotated to genomic regions based on their relation to the nearest CpG island. CpG islands were annotated based on the UCSC Genome Browser (average length = 672 bp in the data; min = 201 bp; max = 15,960 bp). Shore is the flanking region of CpG islands covering 0–2000 bp distant from the CpG island. Shelf is the region flanking island shores covering 2000–4000 bp distant from the CpG island. d A higher percentage of CpG sites are directly disrupted by the SNP in mQTL pairs compared to by chance alone (horizontal dashed line), and more so than in non-mQTL pairs (p value < 2.2 × 10⁻¹⁶). Such enrichment decays with increased FDR thresholds. *p < 0.05, **p < 0.01