qqplot for the P-values dervided from permutation test in Plink

Question

[Shicheng Guo]

Any one tried qq-plot for the P-values derived from permutation test in plink (Both EMP1 and EMP2)?

I get a quite strange qq-plot for the pvalues from permutation test to linear regression (phen ~ geno + age)

I also want to know the relationship between EMP1 and EMP2.

EMP1 > EMP2 or EMP1 < EMP2 or they don't have any relationship?

Thanks.

Here, I give a qq-plot for linear regression (phen ~ geno + age)

[Kevin Blighe]

Please share the QQ plots via, for example, ImgB. How about the non-permuted P values? How have you pre-filtered your SNPs? - genomewide or focused on a particular locus?

Also to add: if you have low sample numbers or your cases/controls are unbalanced, then QQ plots will invariably look odd, or, if you are adjusting too much, they may also look odd.

[Kevin Blighe]

You edited the post, I see - thanks!

EMP1 is the empirical P value whilst EMP2 is the correct P value (after permutation). Your QQ plot for EMP2 looks 'fine'. My thoughts are that the small tail at the top-right of the EMP2 QQ plot may be SNPs that are related to your phenotype of interest (phen).

If the P values fit perfectly to the expected distribution, then, of course, it would indicate that your populations indicated by phen are the same (null hypothesis).

[Shicheng Guo]

Hi Kevin, It's genome-wide association P-values, not selected from any chrosome. The top-right of EMP2 is my interesting SNPs. But 1) I don't understand EMP1, why they are lower in expected quantiles of P-values 2) I don't understand the relationship between EMP1 and EMP2.

Also I don't understand is that for some SNPs, the P-values of EMP2 is smaller than EMP1

[Shicheng Guo]

Yes. Chris told me these EMP1 and EMP2 is only for genetics variants, not for covariants. If I want to show the P-values for co variants I can add --tests

[Shicheng Guo]

Hi Kevin, How to understand the observed P-value in left figure is lower than the expectation regions? and sometimes, the observed P-values in higher than expected regions (null hypothesis). Which factors will caused the observed P-value higher or lower than the expected regions (grey region in the figure) . Thanks.

[Kevin Blighe]

To help to understand that, you may want to take a look at this Stats StakOverflow answer: https://stats.stackexchange.com/questions/101274/how-to-interpret-a-qq-plot

• Above the expected distribution implies that the data is more 'spread' (dispersed) than expected.
• Below the expected distribution implies that the data is more 'compressed' than expected