Almost every ancient-DNA ancestry analysis — ours included — rests on one public dataset: the Allen Ancient DNA Resource, or AADR. It is the reason a consumer product can model a genome against thousands of excavated individuals without sequencing a single bone. This article explains what the resource is, what a version of it contains, and how it becomes the reference behind a report.
What the AADR is#
The AADR is a curated compendium of published ancient human genomes, maintained by David Reich's laboratory at Harvard Medical School and named for the Paul G. Allen Family Foundation, which funded its creation. It gathers genotype data from ancient-DNA papers published by many groups worldwide, reprocesses them through a uniform pipeline, and releases them as one dataset with a single, shared set of marker positions and one annotation table — with each individual's date, find location, skeletal identifier, sex, molecular haplogroups and citation. The resource is described by Mallick and colleagues (2024) in Scientific Data, and its licensing is CC0 for the curated release, which is what makes it usable in products like ours.
Its importance is easy to state: before the AADR, comparing your genome with ancient individuals meant downloading dozens of papers' supplementary files, each in its own format at its own positions, and merging them yourself. After it, there is one file.
What a version contains#
The resource is released in numbered versions as new papers are added. The one our
qpAdm analysis runs against is v66, roughly 23,265 samples across roughly
6,015 distinct population labels. A "sample" is one individual (occasionally one library from an
individual); a "population label" is the group name the publishing authors gave — typically
site, period and culture, such as Turkey_N or Russia_Samara_EBA_Yamnaya.
Two things about those labels matter for reading any result built on them:
- They are the authors' groupings, not natural kinds. Two papers can label similar individuals differently, and one label can hide real substructure. Curating source populations from them — deciding which labels are coherent enough to model against — is analytical work, which is why we publish a curated set of 47 source populations across two eras rather than the raw six thousand. Each has its own page in the ancestry population directory.
- Sample quality varies enormously. The resource includes individuals with millions of covered positions and individuals with tens of thousands. A population average built from three low-coverage samples is a noisy point. Coverage is recorded per sample, and responsible curation reads it.
1240K versus Human Origins#
The AADR is released on two marker sets, and the distinction runs through the whole field.
The 1240K panel is roughly 1.2 million positions chosen for targeted enrichment of ancient
DNA — the set most ancient genomes since 2015 were captured on. It has the most ancient individuals
and is the panel most formal modelling runs on. Our qpAdm merge uses the AADR's wider 2M release
(2,142,271 positions across 23,265 individuals in v66), which contains the 1240K capture positions
together with the Human Origins array positions, so a consumer chip overlaps it at every position it
could overlap either panel at. Your raw file is placed at these positions alongside every ancient
sample, and the number of your markers that overlap is your coverage; no other AADR panel can raise
that number, because your chip, not the panel, sets it. The merge is done with the Poseidon toolchain (trident forge), which is the standard
way to combine EIGENSTRAT-format datasets. The report's model record names the panel the published
model ran on and the exact AADR label of every source, suffix and all, beside its catalog name.
The Human Origins panel is roughly 600,000 positions, chosen to be free of ascertainment bias towards Europeans, and is the panel that includes a large set of modern populations sampled for population-genetic comparison. It is the right panel for PCA and for questions that need present-day references. Our free AdmixTools 2 Lab runs over the AADR Human Origins panel, and the free Ancient Sample Atlas maps every dated, geolocated individual in the v66.p1 Human Origins release.
How the resource becomes a reference#
Turning the AADR into the reference behind a report takes four steps, and each is a place where a product can be more or less honest:
- Choose a version and say which. Results are only reproducible against a named version. Ours states v66 in the report.
- Curate source populations. Decide which labels are coherent, era-appropriate and well enough covered to model against; pool nothing under an umbrella label; give each a description and a date range. This is where "every population in the dataset" would be a mistake — the AADR is a sample of the dead, whoever was excavated and published, not a balanced survey of the past.
- Merge the customer's genotypes at the panel's positions. For a chip export that means a few hundred thousand overlapping markers; for a whole-genome VCF, most of the panel — see Upload a whole-genome VCF.
- Model, and publish the model with the reference it was run against. A weight without its right set and its panel version cannot be checked.
The Model Lab unlock on a published report includes the download of the exact merged EIGENSTRAT bundle, so the whole chain can be reproduced on your own machine — see Run your own qpAdm models.
What the resource is not#
It is not a family tree. An individual in the AADR is a person who died, was excavated, was sequenced and was published; being modelled against a population of such people says your genome is compatible with drawing ancestry from a population like theirs. It never says those particular people were related to you, and no analysis built on the resource can. Sharing a stretch of genome with one of them — the subject of Ancient DNA matches explained — is identity by state, not proof of descent.
It is also not complete or uniform. Regions and periods with active ancient-DNA programmes — Europe, the Near East, the steppe — are densely sampled; much of Africa, South Asia and the Americas is thin. A model's sources can only come from what has been published, which is why a "missing" population in a result is sometimes a missing population in the record.
Reading the annotation table#
If you download the resource yourself, the file worth reading first is the annotation table — one row per individual. The columns that matter for interpretation are the date (given as a mean and a standard deviation in years before 1950, whether from direct radiocarbon dating or from archaeological context — the table says which), the locality and coordinates, the group label, the number of SNPs covered on the 1240K panel, the molecular sex, the Y and mtDNA haplogroups where determinable, and the publication. A population average that pools individuals from a wide date range, or mixes directly dated with context-dated samples, is a different object from one built from a single well-dated cemetery, and the table is where that difference is visible. It is also the source of every dossier fact on our own population pages.
The other reference in the field#
Global25 coordinates come from a different construction: a PCA built by the independent Eurogenes service on its own compilation of ancient and modern samples, including many drawn from the same published papers. It is not the AADR, it is not versioned the same way, and no one but that service produces coordinates in it — see What are Global25 coordinates?. Which reference answers which question is the subject of qpAdm vs Global25; the terms are in the glossary.
References#
- Mallick, S. et al. (2024). The Allen Ancient DNA Resource (AADR): a curated compendium of ancient human genomes. Scientific Data, 11, 182.
- Haak, W. et al. (2015). Massive migration from the steppe was a source for Indo-European languages in Europe. Nature, 522, 207–211.
- Patterson, N. et al. (2012). Ancient admixture in human history. Genetics, 192(3), 1065–1093.
- Schmid, C. et al. (2024). Poseidon — a framework for archaeogenetic human genotype data management. eLife, 13, RP98317.
