Every method has a birthday. qpAdm's is quiet even by the standards of statistical genetics: it first appears in the supplementary information of Haak et al. 2015, the Nature paper that made "massive migration from the steppe" a household phrase in archaeology, described as "new statistical methods that are substantial extensions of a previously reported approach". No methods paper of its own, no name in the abstract. Within five years it was the standard instrument for ancestry decomposition in ancient DNA, the method behind most of the headlines, and the tool a hobbyist can now run on their own genome. This is the story, told for readers who want to understand why the field trusts it, and precisely how far.
What existed before#
The first ancient-genomics decade ran mostly on two kinds of evidence: clustering (PCA and ADMIXTURE bar plots, descriptive by construction) and f-statistics, the formal tests introduced by Patterson et al. 2012, which could demonstrate admixture but not decompose it into proportions with honest uncertainties. Lazaridis et al. 2014 could show that present-day Europeans need at least three ancestral populations; putting defensible percentages with standard errors on each component, per population, at scale, was the missing instrument.
2015: the identity that became a method#
Haak et al.'s supplement (SI 10) states the idea in one line: if a target's ancestry comes from sources in proportions α₁ to αₙ, then every f4-statistic of the target against outgroups equals the same mixture of the sources' f4-statistics. Ancestry proportions become the solution to an over-determined system of linear equations in measured f4 values: solved by least squares, with errors by block jackknife, and a rank test (qpWave, the companion method) asking whether the proposed number of sources is even sufficient relative to the outgroups. The paper used it to put numbers on the Corded Ware culture's steppe ancestry, and the framing of European prehistory as three ancestral populations plus a steppe wave (the frame our steppe ancestry guide works inside) has run through the literature ever since.
What made qpAdm the workhorse was less elegance than fit to the data the field actually has: it tolerates missing data and small samples, works from allele frequencies of pseudohaploid ancient genomes, needs no phasing, and returns exactly the objects an argument needs: weights, standard errors, and a p-value that can reject the model. Lazaridis et al. 2016 standardised the outgroup sets; Skoglund et al. 2017 introduced rotation; and after Reich-lab code was reimplemented as ADMIXTOOLS 2 (Maier et al. 2023), a decade of papers' worth of modelling became runnable on a laptop.
The stress-testing decade#
A method that settles arguments attracts scrutiny, and qpAdm got a proper audit. Harney et al. 2021 validated the machinery on simulations (weights unbiased, p-values uniform under the truth) and quantified the operating limits: ranking models by p-value picks the true one in only 48% of cases; too many outgroups reject correct models; sister-source populations are indistinguishable without a reference that splits them. Williams et al. 2024 measured resolution directly (sources closer than FST ≈ 0.002 cannot be told apart) and priced the usual filters: in their simulations of recent history, p ≥ 0.05 alone gave an 84% false-discovery rate, and adding conditions on the weights cut false positives more than fourfold while bringing the false-discovery rate only to about 57 to 61%.
Flegontova et al. 2025 then showed that how you search matters as much as the statistic, in two different ways. On simulated admixture-graph histories, non-stratified rotating screens reached false-discovery rates of 72.5 to 100%, while temporally stratified (distal) protocols held a fraction of that and improved further when corroborated by independent methods. On simulated stepping-stone landscapes, where groups exchange migrants with their neighbours continuously, the protocols stopped differing significantly; what decided the error rate there was sampling density and the pre-study odds, how plausible the tested models were before any of them ran. And on a landscape, a fitting model can describe where a target sits between sampled groups without any such mixture having happened.
Read one way, that literature is deflating. Read correctly, it is the reason qpAdm results are defensible: the failure modes are published and quantified, most are avoidable by protocol, and the rest tell you how far a result can be read. That is exactly what separates a method from a black box, and what no percentage calculator offers. The audits did not dethrone qpAdm; they wrote its operating manual, including the page on what a fitting model does not prove.
Why it reached consumers at all#
Nothing in qpAdm requires the target to be excavated. A modern genotype file merged into the AADR reference panel is a legitimate target, and a modern target with ancient sources is automatically a distal model. That protects against the class of errors the admixture-graph simulations exposed. It does not, by itself, make a model reliable, because on landscape-like histories the protocol class matters far less than sampling density and the candidates tested.
So qpAdm as a consumer analysis stands on two things. The first is the same statistic and reference data the field uses, with the published discipline (temporal stratification, simplest models first, composite acceptance rather than p-chasing) applied to one customer's file at a time. The second is an honest reading of the result as ancestry composition relative to the sampled groups, not as a dated migration. In our case, candidate sources are chosen for each customer's declared origins before any model runs, and a person reviews the candidates and approves every published model.
A supplement note in 2015; the field's standard by 2018; audited to its edges by 2025; and now a checkable statement an individual can buy about their own deep ancestry, as long as it is read for what it is. Methods rarely age this well, and the ones that do are the ones whose limits got published.
Terms used here are defined in the glossary.
References#
- Haak, W. et al. (2015). Massive migration from the steppe was a source for Indo-European languages in Europe. Nature, 522, 207 to 211. (SI 10: first qpAdm and qpWave use.)
- Patterson, N. et al. (2012). Ancient admixture in human history. Genetics, 192(3), 1065 to 1093.
- Lazaridis, I. et al. (2014). Ancient human genomes suggest three ancestral populations for present-day Europeans. Nature, 513, 409 to 413.
- Lazaridis, I. et al. (2016). Genomic insights into the origin of farming in the ancient Near East. Nature, 536, 419 to 424.
- Harney, É., Patterson, N., Reich, D. & Wakeley, J. (2021). Assessing the performance of qpAdm. Genetics, 217(4), iyaa045.
- Maier, R. et al. (2023). On the limits of fitting complex models of population history to f-statistics. eLife, 12, e85492.
- Williams, M. P. et al. (2024). Testing times: disentangling admixture histories in recent and complex demographies using ancient DNA. Genetics, 228(1), iyae110.
- Flegontova, O. et al. (2025). Performance of qpAdm-based screens for genetic admixture on graph-shaped histories and stepping stone landscapes. Genetics, 230(1), iyaf047.
Editorial note: revised on 24 September 2026. An earlier version reported only the admixture-graph half of the 2025 audit, treated a modern target's distal models as if they sat in a low-error class, and described our own analysis in terms that no longer matched how it runs. This version adds the stepping-stone landscape results and corrects both descriptions.
