Ask where Global25 coordinates come from and the answer is a blog. Not a company, not a university consortium — the Eurogenes blog, written by Davidski, the same pseudonymous author whose K13 and K15 calculators defined GEDmatch-era admixture analysis. The Global25's route from one blogger's experiment to the ancestry community's common currency explains a lot about what the coordinate is and is not, so it is worth telling properly.
Before coordinates: the calculator years#
Eurogenes began in the ADMIXTURE-calculator era, alongside Dodecad and HarappaWorld — that story is here. By the mid-2010s the limits of frozen component calculators were plain to the people building them: components drifted with the training panel, similar names meant different things across projects, and the ancient-DNA flood (2014–2016: the first Neolithic farmer, hunter-gatherer and steppe genome panels at scale) kept making 2012's constructs look dated. The blog's response was to stop shipping components and start shipping positions: a shared PCA space anyone could compute with.
Global 10 arrived in October 2016 — a ten-dimension PCA "genetic map" of world populations, published as a plain datasheet that readers could plug into R scripts like Ger Huijbregts's nMonte to model ancestry proportions themselves. The insight that stuck was less the map than the format: rows of numbers, portable to any tool, with the reference samples' rows published alongside.
2018: Global25#
In February 2018 Davidski announced the successor, extending the space to 25 dimensions — enough to carry fine intra-regional structure worth modelling. The blog ran "Global25 workshop" tutorials through 2018–2019 covering PCA plotting, distance runs and nMonte modelling, and personal coordinates were offered as a paid service (originally around twelve US dollars; today the independent G25 Requests portal issues them at €15 per kit). Two design decisions from those years still govern everything downstream:
- The space is fixed. New samples are projected into the existing 25 axes rather than the PCA being recomputed — which is why your row never changes, why rows from different years are comparable, and why an entire tool ecosystem could standardise on the format.
- Both forms are published. Every row ships scaled and unscaled, and every mix-up between the two produces garbage without an error message. The convention (scaled, almost everywhere) hardened early.
The ecosystem then filled in around the format: VahaduoJS — a free browser tool by a pseudonymous community developer — became the standard way to run distances and mixtures (our tutorial); datasheets of ancient population averages grew with each published paper; and forums settled on G25-plus-nMonte as the lingua franca for arguing about ancestry models.
Why it stuck#
Plenty of PCAs existed in the literature. Global25 won the hobby because of three properties the academic ones never offered together: personal projection (anyone with a raw file could get their own row), published references (the ancient and modern averages to compare against, updated as papers landed), and portability (plain text rows any tool could ingest). It made ancestry modelling reproducible at home — your row plus a stated source panel plus a stated fit is an argument someone else can re-run, which no black-box percentage from a testing company ever was.
The honest counterweight: none of that makes a coordinate a test. A G25 fit cannot reject a model, 25 dimensions compress real structure, and the space itself was built by one person's choices about samples and processing — held to no peer-review process, but also held to a market test the academic tools never face: fifteen years of users checking results against known ancestry.
Where it stands now#
Global25 remains independent. Coordinates come from the Eurogenes service and nowhere else — no testing company issues them, simulated rows are a different object, and we obtain official rows on customers' behalf through the Coordinate Concierge rather than computing anything ourselves. What has kept growing is the analysis layer: era-scoped reference panels (ours: 30,386 samples in 1,535 curated populations across six eras), browser tools for distance, admixture and PCA, and paid analyses like our Global25 report that put a curated panel and a written rationale around the same arithmetic the 2018 workshops taught.
One blogger's datasheet turned out to be the most durable piece of infrastructure amateur population genetics has produced. Knowing its history is the best inoculation against both failure modes it invites — treating the coordinate as corporate-grade truth, or dismissing it as a hobbyist toy. It is neither: it is a shared, fixed, checkable space, and everything good about it follows from those three words.
Terms used here are defined in the glossary.
