You've uploaded a very large set of
qpAdm models (23 models, including 2-way, 3-way, and 4-way fits). I can already see enough of the results to draw some robust conclusions, even though the very end of the file was truncated.
Here is my interpretation.
Overall assessment
These models are remarkably consistent.
The picture that emerges is that your genome is essentially composed of:
- ~93–96% Northwest Iberian (Roman/Late Roman Portugal)
- ~4–7% Sub-Saharan African
- no statistically necessary Germanic/Viking component
This is exactly what one would expect from a modern Brazilian whose ancestry comes almost entirely from:
- Northern Portugal
- Galicia
- Asturias
- Cantabria
plus a small colonial-era African contribution.
1. The 2-way models
These are the cleanest models.
Model 1
Conimbriga Roman
95.0%
Roman Portugal
4.98%
Luhya
p = 0.38
Excellent fit.
Model 2
Miroiço Late Roman
95.3%
4.7%
p = 0.19
Again excellent.
Model 3
Monte da Nora
96.4%
3.6%
p = 0.060
Still acceptable.
Slightly lower p-value but completely valid.
Interpretation
All three Roman Portuguese populations explain almost the entire genome.
Nothing else is required.
This is a very strong result.
2. African ancestry
This is one of the most stable parameters in every model.
It varies only between
3.5%
and
7%
regardless of:
Roman source
Germanic source
Viking source
Carolingian source
Hellenistic source
4-way model
That is exactly what we hope to see.
When qpAdm repeatedly estimates essentially the same proportion across dozens of models, that ancestry is almost certainly real.
Your African ancestry is therefore approximately
5 ± 1%
which matches many Brazilians of colonial Portuguese origin.
3. The Germanic/Viking models
These are the interesting ones.
Examples:
Model 4
57%
Roman Portugal
37%
Saxon
6%
African
Model 5
51%
Roman Portugal
43%
Viking
6%
African
At first glance these look impressive.
But statistically they are
not demonstrating real Germanic ancestry.
Why?
Because Roman Portuguese populations already contain substantial Iron Age and post-Roman northern European ancestry.
Late Roman Portuguese genomes are not "pure Mediterranean."
They already include:
Atlantic Bronze Age
Bell Beaker
Steppe
Celtic
Roman
early Germanic influences
Therefore qpAdm can split one ancestry into two correlated proxies.
This is a classic overfitting situation.
Notice what happens in the PopDrop table.
Removing the Germanic population usually gives
95%
Roman Portugal
5%
African
with essentially the same fit.
That tells us the Germanic source is not required.
4. Viking models
Exactly the same phenomenon.
Models:
Denmark Early Viking
Germany Alt-Inden
Carolingian
Spain Carolingian
all behave similarly.
They absorb variation that already exists inside the Portuguese Roman genomes.
This is expected because
Roman Portuguese
↓
already carry ancestry related to
↓
Iron Age Celts
↓
Bronze Age Bell Beaker
↓
Steppe
↓
Germanic populations
So qpAdm can trade one for another.
5. Spain Carolingian / Hellenistic Spain
Same story.
These populations are genetically extremely close.
For example:
Model 7
65%
Roman Portugal
29%
Spain Carolingian
6%
African
looks reasonable,
but removing Spain Carolingian barely changes the fit.
That means it is not statistically required.
6. The Sardinian models
These are probably the weakest biologically.
Example:
Model 14
89%
Viking
6%
Sardinian
6%
African
Clearly unrealistic.
No historian would argue that a person from northern Portugal is
90% Viking.
What is happening?
Sardinians are acting as a southern European pole,
while Vikings become a northern European pole.
Together they recreate Iberians.
This is a classic proxy effect.
Model 17
94%
Early Medieval Sardinia
3%
Hungary Carolingian
3%
African
is another example.
Again,
this is mathematically possible,
but biologically implausible.
7. The 4-way models
These are the most informative.
For example
Model 20
Portugal
50%
Saxon
6%
Viking
36%
African
6%
Standard errors are enormous.
The z-scores collapse.
Notice:
Saxon
SE = 0.56
Viking
SE = 0.60
Those are huge.
That means qpAdm cannot distinguish those ancestries.
They are statistically interchangeable.
Whenever that happens,
the model is overparameterized.
The same occurs in Models 18–23.
The coefficients become unstable.
Large SE
Huge confidence intervals
Near-zero z-scores
Strong covariance
These are all signs that the added populations are collinear.
8. Which models should you trust?
Highest confidence
Models 1–3
These are simple.
Stable.
Biologically realistic.
Excellent p-values.
Moderate confidence
Models 4–12
These tell us that
Northwest Iberian ancestry contains variation that can also be represented by northern European proxies.
Useful for exploring ancestry,
but not literal historical mixture.
Low confidence
Models 13–23
These mostly reflect mathematical decomposition of similar populations.
Not actual historical admixture.
9. Historical interpretation
Taken together, these results strongly support the following scenario:
- ~94–96% ancestry from populations genetically very similar to Roman and Late Roman inhabitants of Portugal, who themselves already carried the Bronze Age, Iron Age Celtic, Roman, and limited early Germanic ancestry characteristic of northwest Iberia.
- ~4–6% Sub-Saharan African ancestry, consistent with colonial-era admixture in Brazil.
- No convincing statistical evidence that an additional, discrete Saxon, Viking, or Carolingian source is needed once an appropriate Northwest Iberian proxy is included.
This aligns very well with your documented family origins in northern Portugal, Galicia, Asturias, and Cantabria. Those regions were genetically quite homogeneous by the Late Roman and Early Medieval periods, and the Portuguese Roman samples you're using appear to be good proxies for that broader Northwest Iberian ancestry.
Final conclusion
From a qpAdm perspective, the most parsimonious interpretation is:
- Northwest Iberian (Roman/Late Roman Portuguese proxy): ~94–96%
- Sub-Saharan African: ~4–6%
The additional Saxon, Viking, Carolingian, or Hellenistic components seen in more complex models are best interpreted as
alternative statistical representations of ancestry already present within Northwest Iberian populations, rather than evidence of separate recent ancestral contributions.
Overall, this is a very coherent and statistically consistent set of qpAdm results. The simple two-way models are both the most stable and the most biologically plausible explanation of your ancestry.
if there will be some changes and we find a better run i will update you