[apathy]

This paper is a PR gold mine but a methodological disaster, which is par for the course in "EWAS" (epigenome-wide association studies). At one point I had to remind a coauthor that pointing out structural problems in fashionable studies such as these is a quick way to lose (more) faith in modern peer review; for some reason she thought that EWAS enthusiasts gave a flying crap whether the "genes" they were studying were subject to structural variation (e.g. amplification and deletion). Ha! Ha! Ha!

Anyways. Their functional enrichment analysis is uncorrected for the known bias of the platform (something that has been repeatedly addressed by multiple authors since 2012), and no attempt appears to have been made to correct for cryptic stratification (i.e. structural polymorphisms, which are rampant in human populations, and particularly among so-called metabolic genes), though in the study population that may not be a major issue.

Quantile normalization is only appropriate if one can reasonably assert that the overall distribution of measurements is roughly the same between individuals and groups; this assumption has been shown to be invalid in the absence of positive and negative controls for gene expression, whence its original propagation, and more so for DNA methylation under various conditions. The batch correction approach used here is notorious for squashing real signal, although paradoxically that may have moderated some of the other methods choices.

Moreover, the paper demonstrates that a particular sample of high-SES vs. low-SES individuals in Cebu in the Philippines demonstrates some (fairly tiny) differences in DNA methylation at a relatively small number of CpGs (about 2000 out of 485000 or so measured and 110000 or so tested), without particular note as to whether the sites are clustered, functional, or otherwise of interest. The functional impact of these changes are difficult to interpret, partly because of the bias in the functional analysis (something that has been established for nearly a decade; the authors clearly went shopping for methods in a "confirmatory" style).

We shan't even bother to discuss the effect of [mal]nutrition on metabolism and thereby upon DNA methylation and cell composition (both intertwined, although an attempt was made to correct for the interaction), which further muddies the waters w/r/t SES as opposed to individual-level effects. The analysis is done with a fixed-effects model assuming unstructured shrinkage, which of course is a bit odd considering that the measurements have a relatively easily determined correlation structure (their sample size is sufficient to estimate this) and thus variance decomposition could have been highly informative. This is doubly odd for a population "epigenetics" study, given that variance components were literally invented in population genetics.

In conclusion, while it's a lovely piece for a PR department, the actual relevance of either the measurements or the phenomena to actual humans and public policy is quite difficult to interpret. Perhaps that was the point...