What is your biological age? A problem with the epigenetic clock approach.

Having already sent off blood samples to 23andMe and Helix, mainly out of curiosity over possible genome details relevant to my health (Apparently I might be slightly less inclined to dementia issues and more inclined to heart issues), I couldn't resist sending off yet another blood sample to myDNAge, for a test based on Horvath's epigenetic aging clock. The test analyzes DNA methylation patterns of over 2,000 loci on the human genome to give an 'accurate' biological age prediction. I hoped to confirm my smug opinion that my biological age is ~ 10 years less than my chronological age.

I have not yet received a result that might confirm my opinion, but any result I get (after laying out $300 for the test) is now called into question by El Khoury et al. (open source), who show that both the Horvath clock model ,as well as another model by Hannum, systematically underestimate age in tissues from older people. The testing data used in generating this clock did not have a large representation of tissue from elderly individuals, and closer analysis of such date shows a divergence from Horvath's extrapolated line. Details of the analysis are in the open source paper. Here is the summary:

The age prediction properties of both Horvath and Hannum DNA methylation clock models begin to degrade as subjects enter old age. This is at least partly due to saturation, i.e., DNA methylation proportion at some loci approaching 0 or 1, and confounding with the effects of other age-related processes will also play a role. It is likely that this could be ameliorated with additional loci and/or further refined modeling of the currently used set. Association tests using age acceleration should incorporate age as a covariate (as should those using DNA methylation values for individual loci) to avoid spurious associations.

Update on post.....I emailed mydnage.com about this issue, and got the following response:

Dear Deric,

We observed the phenomena mentioned in the recent publication entitled “Systematic underestimation of the epigenetic clock and age acceleration in older subjects” 4 years ago. Our algorithms that are optimized for sample type (optimized either to whole blood or urine) have already took this factor in consideration. In short, your result will not be influenced.