Global25 (G25) for South Asians: What the Coordinates Can and Cannot Tell You

Spend an evening on a South Asian ancestry forum and you will see the same shorthand: pasted G25 coordinates, a Vahaduo screenshot, and a model reading "AASI, Iran Neolithic, Steppe MLBA" with a distance at the bottom. If you have a 23andMe, AncestryDNA or MyHeritage file, the obvious questions are what G25 is, where the coordinates come from, and how far to trust the percentages.

Global25 is a genuinely useful hobbyist tool. It also has limits that matter more for South Asians than for most users, because a key ancestral source for the subcontinent has never been sampled directly.

What Global25 actually is

Global25 is a principal component analysis, or PCA, built and maintained by Davidski of the Eurogenes blog. PCA compresses hundreds of thousands of markers into the few directions along which people differ most, a method long established in population genetics (Patterson, Price and Reich, 2006). What Global25 adds is a large, regularly updated set of ancient and modern reference samples in the same space.

Each sample is described by 25 numbers, one per principal component, PC1 to PC25, and so are you. Because everyone sits in the same space, you can measure how close you are to any reference, plot yourself, or ask which mixture of references lands nearest to you.

Eurogenes publishes reference datasheets in ancient and modern sets, each scaled and unscaled. The blog's own guide says the datasheets are updated "usually every few weeks", that the main purpose of Global25 "is to provide data for mixture modeling", and that scaled coordinates tend to give more stable results. Most hobbyist work uses the scaled set, and the 2018 step-by-step post warns that mixing scaled and unscaled numbers in one model can produce strange results.

How people get G25 coordinates from raw data

There are two separate things people mean by "getting G25".

• The reference datasheets. These are free. Eurogenes links them from the blog, and they are also mirrored for download through the Vahaduo site.
• Your personal coordinates. These are calculated by Eurogenes from your unmodified raw data file, as a paid service. The ordering route has changed more than once over the years, so the Eurogenes blog is the place to check the current instructions, price and turnaround rather than a forum thread from a few years ago.

Some other companies also sell coordinates labelled G25. Before mixing those with official datasheets, ask how they were produced, because coordinates from different pipelines do not always sit in exactly the same space.

Your file must be complete and unedited. If unsure, our raw DNA file check takes five minutes, and what raw DNA data is explains the format.

How hobbyists model G25 with nMonte and Vahaduo

The interesting part is mixture modelling: asking which blend of reference populations best reproduces your position.

1. Pick a target. Usually your own coordinates, or a population average.
2. Pick sources. A short list of reference samples you think stand in for your ancestral streams.
3. Run the fit. The nMonte R script, described in the Eurogenes guide, takes a "data" file of sources and a "target" file and finds the mixture that sits closest to the target. Vahaduo Admixture JS does the same job in a web browser, and works offline once the page is saved.
4. Read the output. You get percentages for each source and a distance: how far the modelled mixture sits from your real coordinates. Lower is closer.
5. Iterate. Swap sources, compare distances, and plot yourself in Vahaduo's PCA views.

This is fast and great for building intuition. It is also where most mistakes happen.

A typical South Asian G25 model

Most South Asian forum models use three ingredients, the structure the peer-reviewed literature found. Datasheet labels change between updates, so treat these as categories, not exact sample names.

Source	What it stands for	Common G25 proxies	Main caveat
AASI	The deep indigenous South Asian hunter-gatherer lineage	Andamanese (Onge); South Indian tribal groups such as Irula or Paniya	No ancient AASI genome exists; every proxy is indirect
Iranian-related	Early Holocene ancestry related to people of the Iranian plateau	Iran Neolithic (Ganj Dareh); Indus Periphery individuals from Shahr-i-Sokhta or Gonur	The South Asian lineage split from Iranian farmers very early
Steppe MLBA	Middle to Late Bronze Age pastoralists of the Central Asian steppe	Sintashta- or Andronovo-associated samples, or similar Central Steppe MLBA groups	Several close variants exist; using two at once invites overfitting

AASI proxies: Onge versus Irula

This is the single biggest decision in a South Asian model. Narasimhan et al. 2019 represented AASI with present-day Andamanese hunter-gatherers, whom they hypothesised are related to the unsampled indigenous South Asians. Onge are distant from mainland AASI but carry no Iranian-related or Steppe ancestry, so they keep the model clean. Irula or Paniya are geographically and genetically closer, but they are modern groups that already carry Iranian-related ancestry. Use them as the "AASI" source and your AASI figure is really "Irula-like" ancestry, a different and larger quantity. Neither choice is wrong; they answer different questions. Our guide to AASI covers the lineage itself.

Iran Neolithic versus Indus Periphery

Plain Iran Neolithic samples such as Ganj Dareh are an imperfect fit for South Asia. The Indus Valley genome from Rakhigarhi carried Iranian-related ancestry from a lineage that, Shinde et al. 2019 found, split from Iranian plateau lineages more than 12,000 years ago. Many hobbyists therefore prefer the Indus Periphery individuals, which already combine Iranian-related and AASI ancestry. That is closer to reality but hides some AASI inside the Iranian-related source. See Indus Valley DNA for the background.

Steppe MLBA

In Narasimhan et al.'s modelling, the immediate source of Steppe ancestry in South Asia is Middle to Late Bronze Age steppe groups (Central_Steppe_MLBA), who themselves carried ancestry from earlier Yamnaya-related pastoralists. The paper found this ancestry reached northern South Asia in the first half of the second millennium BCE and contributes up to 30% of the ancestry of some modern groups. Forum models tend to show it higher in the north-west, with plenty of individual variation. Our post on Steppe ancestry in India covers the range.

How G25 compares with qpAdm in Narasimhan et al. 2019

The South Asian framework that G25 users borrow comes from formal work, chiefly Narasimhan et al. 2019 in Science, which reported genome-wide data from 523 ancient individuals. That paper used qpAdm, a tool in the ADMIXTOOLS package (Patterson et al. 2012) introduced by Haak et al. 2015. qpAdm works on allele frequencies and f-statistics measured against a set of outgroup populations. Crucially, it tests whether a model fits and returns a p-value, so a model can be rejected.

The paper's key South Asian findings, which G25 models try to echo:

• Eleven outlier individuals from Gonur and Shahr-i-Sokhta form what the authors call the Indus Periphery Cline: about 45 to 82% Iranian farmer-related ancestry and about 11 to 50% AASI, with negligible Anatolian farmer-related ancestry.
• Diverse South Asians are fitted as mixtures of three sources: AASI (using Andamanese as a proxy), an Indus Periphery group, and Central_Steppe_MLBA. The present-day groups plotted on the Modern Indian Cline number 140.
• The ASI and ANI both formed largely from Indus Periphery Cline people, with a minimum of about 55% of that ancestry for the ASI and about 70% for the ANI. The idea of ANI and ASI goes back to Reich et al. 2009, which estimated ANI ancestry at 39 to 71% in most Indian groups; see our ANI and ASI explainer.
• Groups that traditionally view themselves as of priestly status tend to have more Steppe ancestry than their ANI/ASI mix alone predicts, with exceptions.

Shinde et al. 2019 then found that the Rakhigarhi individual fits as a genetic clade with the Indus Periphery Cline pool. It is one person, with 31,760 SNPs in the final dataset, so it is evidence rather than a full picture.

	G25 with nMonte or Vahaduo	qpAdm (Narasimhan et al. 2019)
Input	25 PCA coordinates per sample	Genotypes at hundreds of thousands of SNPs
What is fitted	Closest averaged point in PCA space	Shared drift with outgroup populations
Can a model fail?	Always returns an answer; distance is a guide	Yes; a p-value rejects poor models
Review	Hobbyist method, not peer-reviewed	Peer-reviewed, simulation-tested (Harney et al. 2021)

qpAdm is not infallible either: Harney et al. 2021 found it robust in many situations but cautioned against co-analysing ancient and present-day data, very large reference sets, and histories with extended gene flow. Fairly put, G25 explores a hypothesis quickly; qpAdm tests one.

What G25 can and cannot tell you

G25 can:

• Show where your genome sits relative to thousands of ancient and modern references.
• Give a rough sense of your mix of AASI, Iranian-related and Steppe-related ancestry, under the model you chose.
• Let you compare yourself with relatives or published groups quickly and cheaply.

G25 cannot:

• Prove a model. A low distance shows a mixture lands near you; it does not show those were your ancestors.
• Date anything. PCA position carries no information about when mixing happened.
• Tell you your caste, community, religion or identity. DNA shows genetic similarity to reference groups; who you are is a matter of family, culture and choice.
• Give you Y-DNA or mtDNA haplogroups, which are read from specific markers, not PCA coordinates.

The real pitfalls, and how careful users avoid them

Proxy choice

Swap the AASI proxy from Onge to Irula, or Ganj Dareh for an Indus Periphery sample, and the percentages move a long way without anything about you changing. Always report your sources with your numbers.

Overfitting

Add enough closely related sources and almost any target can be fitted with a tiny distance. Davidski's own 2018 step-by-step post says "overfitting should be avoided at all costs", and a 2019 post adds that sources "should always be relatively highly differentiated". Fewer, more distinct sources usually tell you more.

Shifting datasheets

Because the reference set is updated often, the same model can give a different answer next year. Save the datasheet version you used.

Not peer-reviewed

G25 is built on published ancient genomes, but the PCA and its modelling conventions have not been published in a journal. That does not make them wrong. It does mean there is no independently reviewed account of how the coordinates are computed, so forum claims deserve the scrutiny you would give any unreviewed analysis.

Thin or unusual files

Different chips read different markers, and an old file carries less information. See our explainer on DNA imputation for why marker overlap matters.

Already have raw data? Here is the quicker route

If you want a clear South Asian answer rather than a modelling hobby, you do not need coordinates at all.

1. Download your raw file. Step-by-step guides: 23andMe raw data, AncestryDNA raw data and MyHeritage raw data. FamilyTreeDNA and LivingDNA files work too.
2. Leave it unmodified. A .txt or .zip exactly as the company provided it. 23andMe versions 3, 4 and 5 are all accepted.
3. Go to the Helixline upload page and choose a report: Ancestry at ₹1,999 / $25, or Complete (ancestry and health) at ₹3,999 / $50.
4. Pay and upload inside app.helixline.in, where the upload page takes you. Data is stored encrypted in India.
5. Read your report. Results normally arrive within 24 hours of payment and a valid file, showing your closest South Asian reference groups.

None of this stops you using G25 as well: a report for a readable answer, coordinates for experimenting.

G25 or a South Asian report: which suits you?

Choose G25 if you enjoy the modelling itself and are comfortable judging proxies and distances. Choose a South-Asian-focused report if you want an answer you can read in one sitting, without picking sources. For the wider landscape, our comparisons of HarappaWorld on GEDmatch, Illustrative DNA and the best upload sites for South Asian ancestry set out the trade-offs.

Frequently Asked Questions

What are G25 coordinates?

G25, short for Global25, is a principal component analysis (PCA) maintained by the Eurogenes project. Your coordinates are 25 numbers that place your genome in the same compressed space as thousands of ancient and modern reference samples, so you can compare distances and run mixture models.

How do I get my own G25 coordinates?

Personal coordinates are generated by Eurogenes from an unmodified raw data file as a paid service. The ordering route has changed over the years, so check the Eurogenes blog for current instructions and price. The reference datasheets themselves can be downloaded free.

Is Global25 peer-reviewed?

No. Global25 and the hobbyist tools built around it, such as nMonte and Vahaduo, are not published in peer-reviewed journals. They draw on published ancient genomes, but the PCA and the modelling method have not been through formal review the way qpAdm has.

What is a good distance in a G25 model?

Lower means the modelled mix sits closer to you in G25 space, but there is no fixed pass mark. Compare models only on the same target and datasheet, and distrust a very low distance built from many similar sources, which is often overfitting.

Which AASI proxy should South Asians use in G25?

No ancient AASI genome has been sequenced, so every choice is a stand-in. Andamanese samples such as Onge are distant but unmixed; South Indian tribal groups such as Irula or Paniya are closer but already carry Iranian-related ancestry. Report which proxy you used, because it changes the numbers.

Does Helixline give G25 coordinates?

No. Helixline does not produce G25 coordinates. An upload report compares your raw data with South Asian reference populations and shows your closest reference groups and genetic similarity, explained in plain English, usually within 24 hours of payment and a valid file.

Conclusion

Global25 puts thousands of published genomes within reach of anyone with a raw data file, and its standard three-source frame of AASI, Iranian-related and Steppe MLBA ancestry follows the peer-reviewed work closely. Its weakness is that of any flexible tool: the answer depends on the question you ask. Use few, clearly distinct sources and stay suspicious of very low distances. And if what you want is a readable account of how your DNA compares with South Asian populations, the file you already have is enough.

Read next: What is AASI? Ancient Ancestral South Indian DNA explained.