The approach described in previous sections relies on ANOVA to detect differentially methylated CpG sites and takes individual sites as a starting point for interpretation. Since ANOVA compares M values at each site independently, this strategy is robust to type I/type II probe bias.

An alternative could be to first summarize all the probes belonging to a CpG island region (i.e. island, N-shore, N-shelf, S-shore, S-shelf) and then use ANOVA to compare regions across the groups. Since the summarization will include both type I and type II probes, you may want to split the analysis in two branches and analyze type I and type II probes independently.