qpAdm online
Run qpAdm online, with nothing to install
Fit qpAdm ancestry models on the Allen Ancient DNA Resource in your browser. Choose the sources and right populations, use a population or your own DNA file as the target, and read the p-value and the ancestry weights.
What you get
Nexju loads the reference data, computes the statistics and runs the checks.
qpAdm in Nexju
6 featuresThe Allen Ancient DNA Resource, already loaded
Release v66 on its 1240K and Human Origins SNP sets, with more than 23,000 genomes on 1240K and more than 28,000 on Human Origins. Genomes keep the AADR’s own population labels, and genomes from newer papers are added as supplements. You download, convert and merge no genotypes yourself.
Your own genome as the target
Upload a raw DNA file from 23andMe, AncestryDNA, MyHeritage, FTDNA, LivingDNA or tellmeGen. Nexju merges it with the reference data privately. The free plan keeps 5 files.
p-values, weights and standard errors
Each model reports its tail probability, its SNP count and the weight of every source with a block-jackknife standard error. A model is plausible when p ≥ 0.01 and every weight lies between 0 and 1.
Rotations over candidate sources
Swap one source at a time or try every combination from a candidate list, and compare the models that fit side by side.
Checks that a model holds up
On Pro, you can drop right populations, let source sets compete or add a source, and see whether a fitting model still fits.
Checked against ADMIXTOOLS
Nexju’s qpAdm is a native implementation, checked against the classic ADMIXTOOLS qpAdm program on fixed test data.
A qpAdm model in three steps
Every model is a target, a set of sources and a set of right populations.
1. Choose the target
The target is the group whose ancestry the model explains. It can be an AADR population, a set of genomes you pick or your uploaded DNA file.
2. Choose sources and right populations
Pick up to 10 source populations for the target’s ancestry. Then pick more right (outgroup) populations than sources, chosen so that they tell the sources apart.
3. Run, then read the result
Nexju fits the model on its servers. Read the p-value first, then the weights and their standard errors. A rejected model tells you which source set to change.
A worked example, with its numbers
One model that fits and a simpler one that is rejected, as Nexju reported them.
Corded Ware, two models
Human Origins SNP set| Model | Yamnaya | Anatolian farmers | Hunter-gatherers | p-value | Result |
|---|---|---|---|---|---|
| Three sources | 72.5% ± 1.7 | 20.9% ± 1.7 | 6.7% ± 1.7 | 0.014 | Plausible |
| Two sources | 76.2% ± 1.6 | 23.8% ± 1.6 | not in the model | 0.00003 | Rejected |
The target is 55 Corded Ware genomes dated between about 4,900 and 4,200 years ago. The sources are 27 Yamnaya genomes from Samara, 22 early farmers from north-west Anatolia and 4 western hunter-gatherers. The ten right populations are Mbuti, Mota, Ust’-Ishim, Kostenki 14, Goyet Q116-1, Papuan, Karitiana, Natufians, Iberomaurusians and Neolithic Ganj Dareh. Weights are shown with one standard error.
Read the p-value first. The three-source model has p = 0.014 on 128,186 SNPs, above the 0.01 threshold Nexju uses, so it is plausible. It is a narrow pass. At the 0.05 threshold that some papers use, it would be rejected. Every weight lies between 0 and 1, and the smallest, 6.7%, is four standard errors from zero.
Without the hunter-gatherers the model is rejected (p = 0.00003). Yamnaya and Anatolian farmers alone do not account for Corded Ware. The weights of the rejected model still look reasonable, which is why the p-value comes first. The three-source result agrees with Haak et al. (2015), who estimated that Corded Ware people in Germany took about three quarters of their ancestry from the Yamnaya.
References
- Haak, W. et al. Massive migration from the steppe was a source for Indo-European languages in Europe. Nature 522 (2015). doi:10.1038/nature14317
- Harney, É. et al. Assessing the performance of qpAdm. Genetics 217 (2021). doi:10.1093/genetics/iyaa045
- Maier, R. et al. On the limits of fitting complex models of population history to f-statistics. eLife 12 (2023). doi:10.7554/eLife.85492
- Mallick, S. et al. The Allen Ancient DNA Resource (AADR). Scientific Data 11 (2024). doi:10.1038/s41597-024-03031-7
Nexju or ADMIXTOOLS on your own machine
The method is the same. The difference is who prepares the data and the machine.
Three ways to run a qpAdm model
3 ways| What it takes | Nexju | ADMIXTOOLS 2 in R | Classic ADMIXTOOLS |
|---|---|---|---|
| Set-up | Sign in with Google | Install R and the admixtools package | Compile the C programs on Linux |
| Reference data | AADR v66 already loaded | Download the AADR genotype files | Download the AADR genotype files |
| Your own DNA file | Upload it, and Nexju converts and merges it | Convert and merge it yourself | Convert and merge it yourself |
| One model | Pick populations from lists | Call qpadm() in an R script | Write a parameter file |
| Many models | Rotations built in | Script the loop in R | Script the loop in the shell |
| Cost | Free plan, or a paid plan for more runs | Free | Free |
If you already work in R, ADMIXTOOLS 2 is free and flexible. The guide How to run qpAdm shows a model in all three.
Learn the method
Plain-language guides to qpAdm, with worked examples.
qpAdm guides
8 guidesFrequently asked questions
About running qpAdm online, the method, your DNA and the plans.
Can I run qpAdm online for free?
Do I need R, ADMIXTOOLS or a Linux machine?
Can I run qpAdm on my 23andMe or AncestryDNA file?
What p-value counts as a fit?
Is this real qpAdm, or an approximation?
How is qpAdm different from G25 and nMonte percentages?
Which reference data does Nexju use?
Which DNA files can I upload?
How long does a qpAdm run take?
Does a qpAdm result tell me who my ancestors were?
Is my DNA file private?
Can I run qpWave and f-statistics too?
What do Pro and Max add?
Fit your first qpAdm model
Sign in with Google and run a model on the Allen Ancient DNA Resource. The free plan needs no card.