If you have spent any time around amateur ancient-ancestry analysis you have seen a line like this:
Sample,0.121791,0.179749,0.026021,-0.069445,0.074783, … ,0.010418
That is a Global25 coordinate row — G25 for short — and it is the common currency of the whole hobby. Distance calculators take it. Admixture calculators take it. PCA viewers take it. Almost nothing explains what it is. This is the explanation.
What the 25 numbers are#
Global25 is a principal component analysis (PCA) built on a large panel of ancient and modern genomes. PCA takes millions of genotype positions and finds the axes along which the samples vary most; each sample is then described by its position along those axes. Global25 keeps the first 25 axes, so every sample — and every new genome projected into the same space — is a point with 25 coordinates.
Three things follow from that construction:
- A coordinate is a position, not a result. Alone it means nothing. It gains meaning only when compared with other positions in the same space.
- The first dimensions carry the most variation. PC1 and PC2 separate the broadest structure (African versus non-African, East versus West Eurasian); later dimensions capture progressively finer distinctions, down to regional structure within Europe.
- Every downstream calculation is arithmetic over these rows. A "closest populations" list is a sorted table of Euclidean distances; an "admixture" breakdown is a search for the weighted average of reference rows that lands nearest your row.
Where the coordinates come from#
Global25 is produced by one independent service — the Eurogenes Global25 service, run by the author of the Eurogenes blog (how a blog's PCA became the hobby's standard is a story of its own). You send a raw DNA file, you receive your coordinate row. No testing company produces G25 coordinates, and neither do we: Ancestrify never computes coordinates. The practical routes are in How to get your Global25 coordinates, with vendor-specific guides for 23andMe, AncestryDNA, MyHeritage, FamilyTreeDNA and LivingDNA. If you would rather not deal with the provider yourself, our €15 Coordinate Concierge obtains the official row from that service on your behalf, with your consent, and runs the Global25 analysis the moment it arrives.
⚠️ There are tools that simulate or convert a G25 row from other calculators' output. A simulated row is a different object: it describes the conversion, not your genome, and every distance and percentage computed from it inherits that. The free G25 Authenticity Check flags rows that look simulated, edited or rounded.
Scaled versus unscaled#
Every G25 row exists in two forms, and mixing them is the single most common error in the hobby.
The raw PCA output is unscaled: each dimension's numbers reflect how much variation that axis explains, so PC1 spans a far wider range than PC25. In the scaled form, each dimension has been multiplied by a factor that brings the later, finer axes up in weight, so distances are not dominated by the first two or three components.
Which one is "right" depends on the calculation. Scaled coordinates are the convention for distance rankings and admixture fitting, because they let the fine structure count. What is never right is comparing a scaled target against an unscaled reference, or vice versa: the result is a distance list with no meaning, produced without any error message. If you paste coordinates into a tool, use the same form the tool's reference panel is in; ours are scaled throughout, and the reference panels in the free G25 distance calculator match. The full treatment — what the scaling factor does and how to identify a mystery row's form — is in scaled vs unscaled, explained.
What you can compute from a row#
Distance. The Euclidean distance between two rows across all 25 dimensions. Rank a target against a reference panel and you have a "closest populations" list. Our reference set holds 30,386 individual samples grouped into 1,535 curated populations, split across six eras from the Late Bronze Age to the modern day; distances are only comparable within one era's panel.
Admixture. Find the weighted average of several reference rows that lands closest to yours. That is what nMonte-style calculators do, reported with a fit distance — how far the best mixture still sits from your row (what counts as a good one has its own guide). Lower is better, but only up to a point: distance falls mechanically as you add sources, so a low fit with many sources proves nothing. Which sources you offer decides the answer; that is why our reports list every population in the calculator, used and unused, and why we can hand-build a source panel around one customer — see Personalized G25 calculator.
PCA position. Plot your row on two of the 25 axes beside the reference samples and you can see where you fall among ancient and modern individuals. The free G25 PCA viewer does this against curated era views.
Averages. Average several individual rows and you get a population average — which is exactly what a reference "population" in any G25 panel is. The Average G25 tool builds one in the browser.
What a coordinate cannot tell you#
This is the part every G25 tutorial skips.
- There is no p-value. A coordinate fit always returns an answer; it cannot reject a model. If you offer the wrong sources you get confident, wrong percentages. Formal rejection is what qpAdm adds, and the trade-offs are set out in qpAdm vs Global25.
- Distance is not descent. The closest population to your row is the one whose average lies nearest in the PCA space — a statement about similarity, not about who anyone's ancestors were.
- Percentages are model-dependent. Change the source panel and the breakdown changes. A number from one calculator is not comparable with a number from another.
- Fine structure is compressed. 25 dimensions preserve a great deal, but two populations that are distinct in allele-frequency statistics can sit close together in G25 space.
Reading a G25 report#
A well-built report tells you the era, the source panel and its size, every population offered to the model, and the fit distance — so the result can be reproduced and argued with. How to read each of those figures, and the failure modes that make confident results wrong, is the subject of How to read a Global25 report. The terms are defined in the glossary.
Three questions people ask about their row#
Can I compare my row with a friend's? Yes, if both are official and both in the same form. The distance between two individual rows is a similarity, and two siblings will typically sit closer to each other than either sits to any population average — which is a useful sanity check on a new row.
Why do my coordinates look different from my parents' average? Because you are one draw from the mixture your parents represent, not their midpoint. Individual rows scatter around the family's centre; that scatter is exactly why reference populations are averages of many individuals.
Do coordinates change when the reference is updated? Your row is a projection into a fixed space; it does not change. What changes over time is the reference panels you compare it against, as new ancient samples are published — so a distance list from two years ago is not wrong, it is computed against a smaller panel.
Where to start#
If you already have a row, paste it into the free G25 distance calculator — nothing is uploaded, and the panels are the same era populations the paid report is computed against. If you want the full analysis — distances, admixture, PCA and the free Notable Matches lens against 172 published ancient individuals — the Global25 analysis is €29.99. And if you have a raw file but no coordinates yet, the Concierge route above exists precisely so the row you analyse is the official one.
References#
- Patterson, N., Price, A. L. & Reich, D. (2006). Population structure and eigenanalysis. PLoS Genetics, 2(12), e190.
- Novembre, J. et al. (2008). Genes mirror geography within Europe. Nature, 456, 98–101.
- Lazaridis, I. et al. (2014). Ancient human genomes suggest three ancestral populations for present-day Europeans. Nature, 513, 409–413.
- McVean, G. (2009). A genealogical interpretation of principal components analysis. PLoS Genetics, 5(10), e1000686.
