1
Thumbs Up/Down |
Received: 1,250/11 Given: 524/7 |
It’s not based on any ancient component. It’s just based on Baloch genetic drift, but so in fact are other components in GEDmatch calculators. They are also based on genetic drift of modern populations. For example Caucasus is usually based and genetic drift of Georgians.
My point on the previous page was to show that people shouldn’t judge a calculator by whether the modelling or oracles has their own ethnic group but should judge A calculator on whether the admixture percentages make sense
So on the previous page I just proved that admixture percentages of GedrosiaDNA K12 make much more sense than Dodecad K12 even though people like the modelling or oracles of Dodecad K12 better. Simple answer is GedrosiaDNA calculators spreadsheets only have few populations
Accuracy of calculator admixture percentages and oracles are two separate subjects. It’s entirely possible to have a calculator that gives good oracle modeling but the admixture percentages are off for everyone
.
Muzh ba staso la tyaro tsakha ra wubaasu
[IMG][/IMG]
Thumbs Up/Down |
Received: 4,864/123 Given: 2,945/0 |
Maybe there's some trick that I'm missing, but when I've tried to make a projected ADMIXTURE run where modern samples are projected on ancient references, I've gotten results like in the second and fourth graphs below, where most populations get at least 5-10% of every component. The same thing happens when I try to make SNP-level Vahaduo models where modern target populations are modeled using ancient source populations.
In supervised ADMIXTURE, you can select which samples are forced to get 100% of each component, but other samples are still allowed to affect the SNP loadings of the same components. So for example in the first graph below, the EHG_NHG component ended up accounting for Siberian ancestry, because the run included 20 samples from each population, and there were more Siberian samples than EHG_NHG samples, so the loadings of the EHG_NHG component were probably impacted more by the Siberian samples than the EHG_NHG samples.
In projected ADMIXTURE, first the SNP loadings of each component are defined using a set of reference samples, and then the components are used to model projected samples. But projecting ancient samples on modern references seems to give more reasonable results than projecting modern samples on ancient references. I haven't seen any academic paper with an ADMIXTURE run where modern samples are projected on ancient references, but some papers like Saag et al. 2019 have done the reverse.
Thumbs Up/Down |
Received: 1,250/11 Given: 524/7 |
Great work again Komi ! I like your thinking and skills. You just exposed how the GEDmatch calculators are flawed since most are unsupervised and a few are supervised. I’m seeing how Iranians score only 1% E. Asian whereas Palestinians and Yemenis score more than them. Not to mention the 1% E. Asian for Iranians is a joke !
None are a perfect solution but i like the supervised and projected a little better.
The trick you are asking about for others not to affect SNP loadings is in supervised to only have carefully selected reference populations and only a couple of non-reference samples in each run. Of course it’s very time consuming since you’re getting results for only a couple of non-references at a time so you have to do dozens of runs but that’s the only way for non-references to not affect SNP loadings.
Unfortunately GEDmatch calculators were designed the opposite way with most of the time non-references even exceeding references!! You can only imagine how much the nun references have affected the SNP loadings. That’s why you’re seeing the odd results of Gedrosia in Dodecad and other calculators
There are currently 1 users browsing this thread. (0 members and 1 guests)
Bookmarks