Combining my reference population with Xing et al data gets me 3,222 3,161 samples but with only about 23,000 SNPs after LD-pruning. The good thing is that this dataset has 544 South Asian samples …