G25 The Moriopoulos G25 Collection - 2026 Edition

[Duarte]

Although this qpAdm model may pass the statistical test, is not a good model, this estimated admixture is implausible.
Iberians, from whom you inherited most of your genome, don’t have ~50% direct (or indirect) Iran_GanjDareh_N admixture.

The use of a modern Iranian population (Iranian.DG) as an outgroup is incorrect, since modern Iranians are themselves admixed and are not ideal for distinguishing ancient Near Eastern ancestries.

This model is not historically or genetically correct. qpAdm tests only whether the proposed sources can statistically explain the target relative to the chosen outgroups. It does not test if the sources are genetically appropriate and if the model is historically realistic.

A more appropriate model would use populations directly ancestral to the Target. A statistically acceptable model is not necessarily a biologically meaningful one, and careful selection of sources and outgroups is just as important as obtaining a good p-value.​

I believe there is an incompatibility regarding SNP IDs (different formats) between my raw data—which uses standard SNP IDs (e.g., rs12345)—and the archaeogenetic reference dataset (HO - Human Origins), which employs a proprietary SNP ID format. During the file merging process, the `plink --bmerge` command likely excluded SNPs due to strand mismatches (requiring a "flip") or other incompatibilities, resulting in a very low number of remaining SNPs in the merged file. When using the standard qpAdm input script (Target, Left Sources, Right Sources) with a merged dataset—specifically "MADA — Brazil_XXXXXXXXX + AADR v66 HO (PLINK)," which combines my raw data (submitted to MADAI) with the v66 HO (PLINK) dataset—execution halts with a series of errors:
Calculating 0 f4 statistics for block 706 of 710...
Calculating 0 f4 statistics for block 707 of 710...
Calculating 0 f4 statistics for block 708 of 710...
Calculating 0 f4 statistics for block 709 of 710...
Calculating 0 f4 statistics for block 710 of 710...
Error in
Consequently, I can only obtain models that are statistically reliable but contextually (historically) unreliable when running the tool in "Rotation Analysis" mode using all non-missing SNPs on a pairwise population basis.
In short, I end up with randomly generated models that are statistically sound but reveal nothing about historical admixture.
For now, I'm giving up on this little experiment, lol.

 

[Duarte]

Although the number of SNPs from my sample included in the merged dataset file was likely insufficient—which almost certainly compromised the selection of populations truly relevant to me in the admixture models presented by qpAdm, not to mention my complete lack of experience with the tool—I would say that I agree totally with Tautalus. The qpAdm should be used sparingly; certainly, if your personal goal is to assess your ethnic admixture as a combination of modern or ancestral populations, then—beyond the qpAdm result—you will necessarily need to use another tool to evaluate your Euclidean fit based on linear distances derived from a PCA (such as G25 or any other tool you deem reliable). qpAdm simply tells you whether a specific admixture (of the "left" source populations) is mathematically and statistically plausible when compared against external populations that theoretically (and historically) are not part of your genetic background (the "right" sources).

Naturally, other point to consider is that laboratories like the Reich Lab place great value on the preliminary use of qpAdm to generate reliable statistical results—establishing a robust archaeogenetic profile of ancestral populations that aligns with the historical and cultural context of archaeological sites, and which can validate or refute scenarios of migration, admixture, or isolation. Over millennia, mutations and genetic drift have caused modern populations to diverge significantly from ancestral groups, resulting in a substantial Euclidean distance between modern and ancient populations. In the context of modern populatios—which are much closer to one another—plotting components on a PCA for the purpose of calculating fits (distances) and admixture proportions plays a far more significant role.

In short, qpAdm tells you: "Statistically speaking, this is what you are—a mix of the 'left' populations—relative to those others you defined as 'outsiders' (the external 'right' sources)."

Both are auxiliary, complementary tools; their results must be analyzed in light of historical and archaeogenetic contexts.

For example, In the next models qpAdm only says to me:

Regarding the in relation to the outgroup to right and the source populations on the left, your admixture can be this one. Statistically it’s a good model. Just this. Now test these results of admixture in a Euclidean Admixture tool to see how the admixtures behave after plotting the principal components in a PCA.

If the Euclidean admixture result aligns with a mathematically sound qpAdm model, then you are on the right track of your modeling.