Yeah maybe the reason why in my IBS run the French were closer to Finns than to Romanians was that all of the Romanian samples had under 580,000 SNPs out of a maximum of 597,573. When I made another dataset with just Finnish, Romanian, and French samples, and I used `--geno 0` to remove every SNP that was missing from any sample, then the French were closer to Romanians than to Finns.
If you use `--geno 0` so there is no SNP with missing data, then does it matter if you combine samples from different sources? Even though I used `--geno 0`, I was now able to make a dataset with 387,406 SNPs and 568 samples, because I only used samples with at least 588,000 SNPs, and I only selected up to 4 samples from each population based on which samples had the highest SNP count. But even now it's weird how Forest Yukaghirs are so close to the French, and the French are still closer to Finns than to Bulgarians:
https://i.ibb.co/nwzP5QH/1.png