---
title: "qpAdm models for South Asian ancestry: AASI, Indus Periphery and the proxy problem"
description: "South Asia is qpAdm's hardest standard fixture: one ancestral stream has no ancient sample at all. The working recipe — Indus Periphery, steppe MLBA, the Onge-as-AASI-proxy problem — with right sets, failure modes and a worked reading."
canonical: https://www.ancestrify.io/blog/qpadm-models-south-asian-ancestry
date: 2026-09-01T09:32:00+00:00
updated: 2026-09-24
author: "Andi Thomaj"
---

# qpAdm models for South Asian ancestry: AASI, Indus Periphery and the proxy problem

Every region teaches a different qpAdm lesson. [Europe teaches the standard recipe](/blog/qpadm-models-european-ancestry);
South Asia teaches what to do when a founding ancestry **has never been sequenced**. The deepest
stream in every South Asian genome — the lineage geneticists call AASI — has no ancient sample,
and every model of a billion and a half people's ancestry routes through proxies for it. That
constraint, honestly handled, is the whole craft here.

## The streams, and the hole in the record

The working framework (Narasimhan and colleagues 2019, still the reference) writes South Asian
variation as combinations of:

- **AASI** (Ancient Ancestral South Indians) — the subcontinent's deep indigenous
  hunter-gatherer lineage, distantly related to Andamanese islanders. **No ancient genome
  exists.** It is inferred, never sampled.
- **Iranian-plateau-related farmer ancestry** — but *not* proxied by `Iran_GanjDareh_N` naively:
  the lineage in South Asia split from Iranian farmers before agriculture reached the plateau,
  which is why the field's proxy of choice is **Indus Periphery** (IVC-era individuals from
  Gonur and Shahr-i-Sokhta genetically continuous with Rakhigarhi), itself already an
  Iranian-related + AASI mixture.
- **Steppe MLBA** — `Kazakhstan_Sintashta_MLBA`-related ancestry (not Yamnaya EBA: the
  steppe stream that reaches South Asia is the later, farmer-admixed one; substituting EBA
  steppe is the region's classic wrong-era error,
  [temporal logic again](/blog/how-to-choose-qpadm-sources-and-outgroups)).

Modern populations then span the **ANI–ASI cline**: ASI ≈ AASI + Indus-Periphery-related; ANI ≈
Indus-Periphery-related + steppe MLBA. The 2009 f4-ratio estimate of 39–71% ANI across the cline
[remains the sanity anchor](/blog/f4-ratio-ancestry-estimation).

## The standard recipe

**Left:** target + `Indus_Periphery` (pooled or West-cluster), `Kazakhstan_Sintashta_MLBA` (or
`Central_Steppe_MLBA`), and an AASI proxy — in practice `Onge.DG`, with the caveat that owns the
next section.

**Right:** [the O9 spine](/blog/how-to-choose-qpadm-sources-and-outgroups#the-standard-right-sets) *minus Onge when Onge is a
source* (never both sides — the cladality rule), plus the contrasts that split Iranian-related
from steppe from AASI: `Russia_EHG`, `Georgia_CHG`, `Anatolia_N`, `Iran_GanjDareh_N`,
`WSHG`/`Botai`-related for the inner-Asian edge, `Mota.DG`, `Ust_Ishim.DG`, `MA1`. Iran_N moves
to the right *because* Indus Periphery carries the Iranian-related stream on the left — the
right-set member that discriminates your sources is
[worth five generic ones](/blog/qpadm-best-practices).

Run [lowest rank first](/blog/qpwave-explained): many southern targets pass as two-source
Indus_Periphery + Onge; northern and upper-caste targets typically need a third, steppe source.
Testing the lowest rank first is statistical parsimony: the simplest model that fits, relative to
the right set used. It is not a head count of the migrations in anyone's history.

## The proxy problem, stated honestly

Onge are **not** AASI. They are a sister lineage separated by tens of millennia of island
isolation and drift — the *least bad sampled relative*, not the ancestor. Consequences, in
descending order of pain: absolute AASI percentages shift by proxy choice (models swapping Onge
for other proxies move weights by real margins, so quote AASI fractions as proxy-conditional);
drift accumulated on the Onge branch can push
[p-values down for reasons that are not model failure](/blog/why-qpadm-models-get-rejected); and
no right set fully [rescues a source that is itself off-topology](/blog/is-qpadm-still-reliable).
The published work handles this with explicit proxy sensitivity checks — running the same model
across proxy choices and reporting the spread. Do the same, or read others' absolute percentages
with that spread in mind.

## The named failure modes

- **Wrong-era steppe.** Yamnaya EBA instead of Sintashta/Central MLBA — passes sometimes, means
  the wrong thing always.
- **Onge on both sides.** Source *and* O9 outgroup — instant
  [entanglement violation](/blog/how-to-choose-qpadm-sources-and-outgroups); prune the right set.
- **East Asian edges.** Munda-speaking and northeastern targets carry East/Southeast
  Asian-related ancestry the three-source model lacks; relative to a right set that can see it,
  [the rank test will usually call for a fourth component](/blog/qpwave-explained): give it
  `China_YR_LN`- or Austroasiatic-associated proxies rather than torturing the right set.
- **Endogamy noise.** Strong founder events in many jati groups inflate drift;
  [SEs widen and p-values roughen](/blog/how-many-snps-does-qpadm-need) even when the model is
  structurally right. More target individuals help more than more outgroups.

## A worked reading

A northwest-Indian-ancestry target: p = 0.18, Indus_Periphery 0.61 ± 0.030, Steppe_MLBA
0.24 ± 0.019, Onge 0.15 ± 0.026; two-source nested models
[rejected](/blog/qpadm-model-record-explained); ~640k SNPs. Reading: compatible, and relative to
this right set and this candidate pool the simpler models do not fit (a statement about the model,
not a count of migrations); the steppe weight is solid ANI-cline-upper territory; and the honest
sentence for the report is "15% AASI *as proxied by Onge*" — with the proxy-swap spread quoted
beside it if the number will bear weight. That sentence is the region's entire lesson in
miniature: [the method is exact about what it tested](/qpadm), and
the analyst's job is to keep the words as exact as the arithmetic.

Terms used here are defined in the [glossary](/glossary).

## References

- Narasimhan, V. M. et al. (2019). The formation of human populations in South and Central Asia.
  *Science*, 365, eaat7487. (The framework, Indus Periphery, steppe MLBA.)
- Shinde, V. et al. (2019). An ancient Harappan genome lacks ancestry from steppe pastoralists or
  Iranian farmers. *Cell*, 179, 729–735. (Rakhigarhi and the pre-agricultural Iranian split.)
- Reich, D., Thangaraj, K., Patterson, N., Price, A. L. & Singh, L. (2009). Reconstructing Indian
  population history. *Nature*, 461, 489–494. (ANI/ASI.)
- Moorjani, P. et al. (2013). Genetic evidence for recent population mixture in India. *American
  Journal of Human Genetics*, 93(3), 422–438. (Dating the cline's formation.)