Regarding G25 PCA, perhaps a way of "judging" it is comparing its results to those from researchers?
By the way, afaik the author of the G25 PCA himself sometimes uses qpAdm to reinforce certain results, to see if the different tools "agree" in something, and researchers may do the same.
From the paper
Genetic origins of the Minoans and Mycenaeans (if you guys know about a better one for this purpose, please tell me), from 2017 - as the G25 PCA release, if I'm not mistaken:
https://static-content.springer.com...jects/41586_2017_BFnature23310_MOESM1_ESM.pdf
"In this section, we model these populations using the qpWave/qpAdm framework1 which allows us to model a Left set consisting of a Test population (whose history of admixture we are investigating) and Reference populations (potential sources of ancestry) in relation to a Right set of outgroups.
(...)
We test for rank=N-1 using qpWave and estimate mixture proportions using qpAdm1 . We report only feasible mixture proportions (in interval [0, 1]). We use a significance level of p=0.05 for rejecting models and mark p-values greater than 0.05 (that represent feasible models) in red.
(...)
We test the robustness of the qpAdm estimates by plotting populations with the inferred mixture proportions (Table S2.26) in the weighted average position of their source populations in PCA space. The results (Extended Data Fig. 5) indicate a close correspondence between the two in all models considered.
(...)
The population from Bronze Age southwestern Anatolia does not form a clade with any single (N=1) population of the All set (p-value for rank=0 < 1e-25). It cannot be modelled as any 2-way mixture (Table S2.8), with the best ones involving a mixture of Anatolian Neolithic and either Iran Neolithic or Caucasus hunter-gatherers.
This population can be modelled as a 3-way mixture (Table S2.9) of ~62% Neolithic Anatolian, ~32% Caucasus hunter-gatherer (CHG), and ~6% Levantine Neolithic ancestry. This extra Levantine Neolithic ancestry parallels the PCA (Fig. 1b) that shows that the Bronze Age Anatolian sample is to the “east” (towards the Levant) relative to the Minoans and Mycenaeans."
So, according to this tool, the best 3-way mixture for SW Anatolian BA is ~62% Neolithic Anatolian, ~32% CHG and ~6% Levantine Neolithic.
I checked G25 results using R, and supposedly using the same components.
The target is the average for SW Anatolia EBA (Isparta).
Using individuals as source / Using Averages
Scaled and penalty = 0 (default in Vahaduo)
TUR_Barcin_N,53 / 53.4
GEO_CHG,28.6 / 29.6
Levant_PPNB,18.4 / 17
Scaled and penalty = 0.001 (as some people prefer)
TUR_Barcin_N,71.4 / 53.8
GEO_CHG,18.2 / 29
Levant_N,10.4 / 17.2
Unscaled and penalty = 0
Anatolia_Barcin_N,52.2 / 54.6
GEO_CHG,25.8 / 25.2
Levant_PPNB,22 / 20.2
Unscaled and penalty = 0.001
Anatolia_Barcin_N,72.2 / 55
Levant_PPNB,15.2 / 20.2
GEO_CHG,12.6 / 24.8
Substantial differences between the results using individuals (rather than averages) and different penalties.
Vahaduo scaled would return 17-18.4% for Levant_PPNB in SW Anatolia EBA average then. Considering samples from Isparta individually as target (rather than the average), the variation would be something like 14-20.6 in Vahaduo scaled (pen = 0), against those 6% from qpAdm.
Conclusion is: The reference samples are supposedly the same, but for whatever reason there's no agreement between that qpAdm's and G25's regarding proportions. If we use researchers' tools as reference, the Levant Neo contribution for SW Anatolia EBA is possibly overestimated by G25. Whilst using pen = 0.001 may decrease the difference in Levant Neo %, it'd affect negatively the other two, especially CHG.