Figure 2: Contribution of different functional classes of variables to DNA methylation variation.

(a,b)Relative contributions of the variable classes according to Random Forest average variable importance (a) and Elastic Net average variable usage (b) are shown averaged across all cancers (Methods). The y axis shows the average rank of each class across cancers (with higher values corresponding to higher contribution). (Boxes extend from 25th to 75th percentiles, centre lines represent the median and whiskers show the minimum and maximum values in each class with the exception of individual outliers shown). (c) Diagram summarizing the steps taken towards calculating overall contribution of each of the met cycle variables relative to other variables in explaining variability in local DNA methylations. (d) Ranking all variables according to their overall selection rate (usage) across all models of local DNA methylation in each cancer. The y axis shows the per cent of variables that ranked lower than each of the met cycle variables (that is, made less contribution to DNA methylation) in each cancer (BHMT2 was removed from the models of colon and bladder cancers due to low expression).