Question: Large Genomic Inflation Factor In Gwas
gravatar for Liyf
8.6 years ago by
Liyf290 wrote:

I use eigenstrat to correct population structure in my multiple GWAS. I want to calculate the genomic inflation factor before correction, I think GIF should be large, for population structure, but my GIF is too large, is 2.14! I never see so large GIF, even before correction, I do not know this may be something wrong with my work?

gwas statistics • 6.5k views
ADD COMMENTlink modified 8.5 years ago by jvijai1.2k • written 8.6 years ago by Liyf290

There are many reasons/artifacts that can be causing this. Starting from QC to the underlying genetics of your phenotype and the variants being tested. Are you using publicly availabe controls? What is the sample size? What kind of population structure do you have? It is hard to answer this question as there are many factors that can cause this.

ADD REPLYlink written 8.6 years ago by Adrian Cortes520
gravatar for David Quigley
8.6 years ago by
David Quigley11k
San Francisco
David Quigley11k wrote:

A large GIF may mean you have made a calculation error, or it may mean you have a P value distribution that doesn't fit the normal GWA model for some reason. It could be correctly indicating that your data violate the population sub-structure assumptions normally made for a GWAS. If you "think the GIF should be large", then what makes you think that 2.14 is "too large"? I think substructure is only one form of bias that could cause a large GIF, though my understanding is it's the most typical cause. If you do a PCA on your genotypes, is there obvious structure, and can you relate that back to ethnicity or region?

You don't tend to see large GIF values in publications because a large GIF indicates there's something about the data that needs to be transformed or filtered before you can do the GWA analysis.

ADD COMMENTlink written 8.6 years ago by David Quigley11k

yes, when I use PCA to correct, the GIF is 1.02. What I am confused is that before correction, GIF too large is acceptable?

ADD REPLYlink written 8.6 years ago by Liyf290
gravatar for jvijai
8.6 years ago by
United States
jvijai1.2k wrote:

It is possible that you may still have SNPs that are not filtered for GWAS QC levels
Try 99% call rate and HWE >10e-06
When you use logistic regression, you may also want to use maybe 3-4 PCAs as covariates.

ADD COMMENTlink written 8.6 years ago by jvijai1.2k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 986 users visited in the last hour