General Population Stratification Discussion

ISG Workshop 2026 Day 4
1

Is it ever the case that one population might have real genetic differences that explain variance in a complex trait, and if so would adjusting for population stratification remove this signal? Or would you just need a different method to look at something like this?

population statification is actually a confounder, so controlling for confouder would reduce the effect size, but true signal should remain (when you have enough sample)

Okay cool, thanks :slightly_smiling_face:

Yes this could be possible - I don’t know if we have any good examples we can think of though sorry.

Maybe the impact of EDAR vs Tricohilin on hair curliness

Perhaps it would be best to do the GWAS within that population only instead?

It is indeed possible, especially if you haven’t removed long-range LD regions in which case PCs can capture e.g. inversions that could be true signal

2

This is the case with height, and unfortunately you do need to control for it and will lose some real signal - it’s a trade-off to reduce false positives by increasing false negatives.

Here is a recent paper that investigates the extent to which PCs actually correlate with PCs due to real effects: https://www.pnas.org/doi/10.1073/pnas.2525790123