I'm using 1000 genome data on European populations. I know there should be some population structure, but smartpca (from eigensoft version 6) gives very different PCs and eigenvalues for each of the 22 chromosomes. By this I mean if I look at the PCA plots, they look very different (some are really separated out by PC1 and 2 whereas others aren't). I have done HWE, missingness, maf>0.01 filtering on this dataset and asked smartpca to remove outliers, so I think the data is very well filtered. The only thing I can think of is that the sample size is so small (I'm using 450 samples) that only a few datapoints/informative SNPs from each population is present in each chromosome, and that changes the PCs. Has anyone else had this problem? If so, did you end up doing smartpca on the whole genome? Or what did you use as PCs to correct for population structure?
read on local vs global ancestry. Apply LD pruning. Use higher maf, be sure to have the final data set of SNPs at least above 10-20k SNPs
Try doing by continent rather than just europe
Thanks. I did LD pruning and I am interested in the local population structure in europe. You might be right in that the problem might be in the maf threshold is set too low.