I am doing a gwas with over 500 plant samples and approx 10 m SNPs obtained after pruning the data with --indep-pairwise 10kb 1 0.8 and --indep-pairwise 50 1 0.8. Then did kinship analysis using pruned data. Tried wald test and lrr test with gemma, with and without population structure as covariates, with and without PCs as covariates, keeping kinship in each case. Also tried lmm with emmax which uses variance components to account for strucure. Tried with both original and sqrt-transformed variables where normality was an issue. With very minor differences, all of my results are consistent and similar. However, the qq plots in each case look bumpy, I add one for example. What could be the reason? How could I improve the analysis - or is it the best I could get?
A QQ plot shows how far your distribution of data compares to a normal distribution. An upwards deviation (inflation) on the right side shows you have identified more low p-values that expected when comparing to a normal distribution. Have you corrected your p-values for multiple testing?
This appears to be a pretty common problem in GWAS analysis. Here is an interesting paper discussing some important considerations.