Abstract
Genetic clustering algorithms, implemented in popular programs such as STRUCTURE and ADMIXTURE, have been used extensively in the characterisation of individuals and populations based on genetic data. A successful example is the reconstruction of the genetic history of African Americans who are a product of recent admixture between highly differentiated populations. Histories can also be reconstructed using the same procedure for groups which do not have admixture in their recent history, where recent genetic drift is strong or that deviate in other ways from the underlying inference model. Unfortunately, such histories can be misleading. We have implemented an approach (badMIXTURE) to assess the goodness of fit of the model using the ancestry “palettes” estimated by CHROMOPAINTER and apply it to both simulated data and real case studies. Combining these complementary analyses with additional methods that are designed to test specific hypotheses allows a richer and more robust analysis of recent demographic history based on genetic data.
The Problem of Over-Interpretation
In modern population genetics, colorful bar plots generated by programs like STRUCTURE and ADMIXTURE have become the standard for visualizing genetic ancestry. These plots represent an individual’s genome as a mixture of $K$ ancestral populations. However, the visual simplicity of these plots often masks a fundamental issue: the "admixture model" relies on a set of rigid assumptions that may not reflect the true demographic history of the samples.
The algorithm is designed to be helpful, finding the most parsimonious way to explain genetic variation. This means it will produce a clean admixture plot even if the underlying history involves severe bottlenecks, sampling biases, or "ghost" populations (unsampled groups). Because qualitatively different histories can produce identical-looking bar plots, researchers often inadvertently construct "just-so" stories about migrations and ancestral origins that the data does not actually support.
Methods: The badMIXTURE Approach
To address this, the authors propose a "goodness-of-fit" approach implemented in a tool called badMIXTURE. The core idea is to move beyond simple allele frequencies and look at DNA sharing patterns using CHROMOPAINTER to create "ancestry palettes."
If the ADMIXTURE results are accurate, an individual’s DNA sharing palette should be a predictable weighted average of the palettes of their inferred ancestral sources. badMIXTURE calculates the "residuals"—the difference between the observed DNA sharing and what the model predicts. Small, random residuals suggest a good fit, while strong, systematic patterns indicate that the "admixture" signal might actually be a result of genetic drift or isolation.
Key Insights and Case Studies
The paper validates this approach across several scenarios:
- Bottlenecks Mimic Admixture: A population that has undergone a severe reduction in size (bottleneck) often appears as an "admixture" of its neighbors. In the case of the Ethiopian Ari,
badMIXTURErevealed that a "unique" ancestral component in the Blacksmith group was actually a signal of recent genetic drift caused by social marginalization, not an ancient remnant population. - The "Winner’s History" and Sampling Bias: The algorithm prioritizes explaining variation in large or highly drifted groups. If one group is over-sampled, others may be forced to look like mixtures of that group, even if their histories are independent.
- Correcting Indian Ancestry: Analysis of 19 Indian groups showed that some, like the Paniya, were inferred as "pure" representatives of Ancestral South Indians primarily because of high levels of recent genetic drift. Once corrected for this drift, their ancestry profiles aligned more closely with neighboring populations.
Conclusion
STRUCTURE and ADMIXTURE plots should be viewed as parsimonious summaries of variation, not as literal maps of ancestral populations that once existed in a "pure" state. While badMIXTURE is not a "truth machine" that automatically reveals the correct history, it provides a vital diagnostic tool to flag when a model is a poor fit. The authors recommend that these clustering results always be paired with complementary methods, such as $f$-statistics or TreeMix, to test specific demographic hypotheses and ensure a more robust interpretation of human history.
Summary updated