Abstract
Inferring the history of population admixture is a cornerstone of modern population genetics. A significant challenge in methods based on Admixture Linkage Disequilibrium (ALD) is the interference of Source Linkage Disequilibrium (SLD)—genetic correlations inherited directly from ancestral populations. Traditional methods attempt to mitigate SLD by ignoring data at short genetic distances and fitting exponential decay curves, but this often leads to biased results in complex scenarios. In this study, we introduce iMAAPs (Inference of Multiple-wave Admixture by fitting ALD using Polynomial spectrum). By formally defining SLD and utilizing a polynomial spectrum (p-spectrum) approach, iMAAPs effectively separates ancestral noise from the admixture signal. Our method demonstrates superior performance in simulated data and provides refined timelines for the admixture histories of the African American (ASW), Maasai (MKK), and Uygur populations.
Introduction
When two or more previously isolated populations meet and interbreed, the resulting "admixture" creates a mosaic of genetic segments. Over time, biological recombination breaks these segments into smaller fragments. By measuring the decay of Admixture Linkage Disequilibrium (ALD)—the non-random association of alleles from different ancestral sources—geneticists can estimate the timing and nature of these mixing events.
However, the accuracy of these estimates is often compromised by Source Linkage Disequilibrium (SLD). SLD represents the background genetic correlations already present in the source populations before admixture occurred. Current tools, such as ALDER or ROLLOFF, typically address this by excluding data below a certain "starting distance" (e.g., < 0.5 cM). This heuristic approach is often insufficient, especially when dealing with multiple waves of migration or continuous gene flow, where the signals of ancient and recent events overlap.
Methodology: The iMAAPs Framework
To overcome the limitations of exponential curve fitting, we developed iMAAPs. This framework rests on three methodological innovations:
1. Formalizing SLD
We mathematically defined SLD within the weighted LD statistic under a two-way admixture model. This allows the background noise to be treated as a quantifiable variable rather than an unknown bias to be avoided.
2. The Polynomial Spectrum (p-spectrum)
Drawing on the Weierstrass approximation theorem, which posits that any continuous function can be approximated by a polynomial, we represent the decay of genetic correlations as a "spectrum." Instead of searching for a single best-fit exponential curve, iMAAPs decomposes the LD decay into a spectrum of potential signals. This allows the model to distinguish between the high-degree signals characteristic of SLD and the lower-degree signals produced by recent admixture events.
3. Reference-Based Denoising
iMAAPs utilizes modern reference populations as proxies for ancestral sources. By calculating the weighted LD in these references, the tool can estimate the SLD and subtract it from the admixed population’s data. This denoising process leaves a significantly cleaner signal for dating the actual admixture events.
Results and Discussion
We validated iMAAPs using both simulated datasets (representing hybrid isolation, two-wave pulses, and continuous gene flow) and empirical data from the HapMap Project and the Human Genome Diversity Project (HGDP).
Empirical Findings
- African Americans (ASW): Previous methods produced highly variable results for the ASW population, with some estimates suggesting admixture as far back as 77 generations ago due to uncorrected SLD. iMAAPs provided a stable estimate of approximately 5.4 generations ago, successfully filtering out ancestral noise that biased other models.
- The Maasai (MKK): Our analysis confirmed a complex, multi-wave history for the MKK. We identified at least two distinct pulses of admixture occurring roughly 16 and 68 generations ago.
- Uygurs: The iMAAPs analysis resolved a major admixture event at approximately 33 generations ago. This finding aligns closely with historical records of gene flow between East Asian and West Eurasian populations along the Silk Road.
The "Starting Distance" Fallacy
Our research highlights that the common practice of adjusting the "starting distance" to avoid SLD is fundamentally flawed. Residual ancestral noise often persists across various distances, creating "ghost" signals that mimic ancient admixture. By modeling the SLD directly, iMAAPs avoids these artifacts, providing a more robust timeline of human migration.
Conclusion and Future Directions
iMAAPs represents a significant advancement in our ability to untangle complex population histories. By shifting from simple exponential fitting to a polynomial spectrum approach, we can more accurately distinguish between single-wave, multi-wave, and continuous admixture scenarios.
While currently optimized for two-way admixture models, future iterations of iMAAPs will aim to address N-way admixtures involving three or more source groups. Additionally, while the tool significantly improves dating accuracy, signals exceeding 500 generations remain a challenge as the mathematical signatures of SLD and ancient ALD converge. Further refinement of the p-spectrum will be necessary to probe these deeper evolutionary timescales.
Summary updated