Weird QQ plot in GWAS analysis

Question

[Đạt]

Dear user, Could anyone can explain my situation. I have microarray data of 53 cases and 53 controls. I conducted logistic regression with age as a covariate. I conducted with typed SNP and imputed SNP The QQ plot of typed SNP is ok. However, the QQ plot with imputed SNP looks like this and many imputed variants have P-value of around 0.2, This is repeated in all 22 chromosomes. I use shapeIT for pre-phase and IMPUTE2 for phasing. Then use the dosage data for analysis. I tried to filter info score with 0.3 , 0.4 and 0.5 and MAF >= 0.05 and 0.1 but it still same.Does anyone face this situation before? Please let me know your advice. Thank you very much

[LauferVA]

did you check your imputation accuracy for each chromosome

[Đạt]

Yes, the concordance range from 0.9 to 0.99, typically at 0.95. Before I exclude those with info score < 0.3 and MAF <= 0.05 it is even like this

Answer 1

[LauferVA]

well, essentially i think this study lacks statistical power. with 53 cases and 53 controls, you would need a very large effect size to see anything at all.

The bump at the beginning is because there aree a huge number of variants that have low MAF, and you may only have a few of them in each category (case and controls), so the p-values are all near 1. That explains the first part.

At the top right, you fall off the O = E line once again because, with only N=53 of each, no variant will achieve a p-value of p < 10^-5 unless the effect size is massive.

Bottom line is, while there could be QC issues lurking here, it is also possible that the quality control is good, and the study lacks power.

I would take the points with highest p-value, start there, and maybe you will find a direction to go.