Fig. 3

Variance captured in complex traits by all available probes and four subsets of decreasing size. Restricted maximum likelihood was used to estimate variance components in the training sample (n ≤ 4450, OSCA software). The four traits (out of seventeen biochemical and complex traits) with the highest proportion of variance captured by DNAm are shown. Five different sets of probes were compared. ‘All available probes’ denotes probes that were common to the Illumina EPIC and 450K arrays and passed quality control procedures in the training sample within Generation Scotland (n = 393,654 probes). The ‘variable non-mQTL probes’ set consisted of probes without reported non-genetic influences and mean Beta-values between 10 and 90%. The remaining three probe subsets contained the 50,000, 20,000 and 10,000 most variable non-mQTL probes (ranked by their standard deviations). The five sets of probes therefore had decreasing numbers of probes but increasing mean variabilities. Vertical bars show 95% confidence intervals. DNAm, DNA methylation; mQTL, methylation quantitative trait locus; OSCA, OmicS data-based complex trait analysis