|
Post by javiermr on Oct 19, 2016 13:41:38 GMT
Hi,
I was wondering what would be the best way to estimate the heritability of a trait in an admixed population. Will the inclusion of PCs (or equivalent) in the mixed models as fixed effects be a possible solution? How could you test if the admixture nature of your sample is affecting the heritability estimates?
Thank you.
UPDATE:
I found this paper: Genome partitioning of genetic variation for complex traits using common SNPs (http://www.nature.com/ng/journal/v43/n6/full/ng.823.html)
They have a section on population structure and if I understood correctly, the idea is the following:
- Population structure causes correlations of SNPs on different chromosomes. Therefore, if you fit the whole genome (or each chromosome separately) you will also capture some of the variance by the other chromosomes and your h2 will biased upwards. - If you fit the chromosomes jointly, you will "protect" against this chromosomal correlations, because the estimate of your variance for each chromosome will be conditional on the other chromosomes.
If this is correct can I:
- Fit the chromosomes jointly and then just add up the h2 of each chromosome to get a h2 accounting for population structure? - What is the difference between estimating the variance using the whole genome and using PCs if I know they capture most of the population structure?
Thanks again.
|
|
|
Post by Jian Yang on Oct 24, 2016 23:49:01 GMT
In the genome partitioning paper (Yang et al. 2011 NG), we utilised the difference in per-chromosome h2 estimate between separate and joint analyses to partition the variance explained by relatedness and population stratification. However, it doesn't mean that the sum of per-chromosome h2 estimates is free of confounding from population structure. To account for population stratification, you might need to fit the first PCs. To account for relatedness, you will need to remove close relatives or fit a model of both SNP and pedigree GRMs (see gcta.freeforums.net/thread/210/use-gcta-greml-family-data).
|
|
|
Post by javiermr on Oct 27, 2016 15:47:29 GMT
From what I understood from the paper (last paragraph of the Quantifying population structure section), you mention that you can either fit each chromosome separately including PCs or fit all chromosomes simultaneously (joint analysis without PCs) and that this approach will lead to similar estimates of the h2 for each chromosome. So, after using either of these approaches, can I then add up all per-chromosome h2 estimates to get the total h2 of the genome, that will be accounting for population structure?
Thank you again.
|
|