How big does big genetics need to be?

In the comments to a previous post defending big genetics, Andro Hsu relates an anecdote that warrants repeating:> IIRC, at the December NIH/CDC meeting Francis Collins suggested that the way to get to the bottom of the missing heritability, the common disease common variant hypothesis, gene-gene and gene-environment interactions, etc. etc. is to run a population-wide, 20-year longitudinal study in which genome-wide data and detailed environmental and behavioral minutiae were tracked for 100,000 participants.

The follow-up commenters starting with John Ioannidis each upped the sample size by an order of magnitude, until someone suggested that the entire U.S. population be sequenced, which it was then realized would require universal health care.

At that point, the meeting ended.

The scary part is that even a study that large might not be large enough: under certain models for the genetic architecture of complex traits (e.g. thousands of common risk variants with tiny effect sizes and substantial interaction between them) the whole US population (or even the entire world's population) might not be large enough to capture the missing heritability, especially given the challenges introduced by sample heterogeneity and measurement error in a study of that magnitude. Subscribe to Genetic Future. Follow Daniel on Twitter.