Evaluating the performance of demographic inference methods
Evaluating the performance of demographic inference methods
Embedded Files
The discovery in the late 1980s that DNA could be reliably extracted and sequenced from ancient human remains—termed "ancient DNA"—revolutionized our capacity to understand human history and the intricate interplay of biological, demographic, and cultural factors shaping contemporary genetic landscapes. Complementing these molecular advances, researchers developed statistical methodologies to translate genetic data into demographic models that could reveal the existence of ghost populations, reconstruct admixture events, and date population divergences. Despite their revolutionary impact, the accuracy of these methods in reconstructing complex population histories remains uncertain, particularly when examining ancient societies characterized by high mobility, genetic similarity, and cultural complexity—challenges further compounded by the degraded nature of ancient DNA. Our laboratory addresses these methodological limitations through simulation-based approaches that bridge critical gaps between genomic analysis and historical interpretation, thereby elucidating when genetic evidence can reliably resolve questions surrounding historical population movements. By establishing methodological guidelines and advancing analytical techniques, our work contributes to more accurate reconstructions of human history through interdisciplinary research that synthesizes ancient DNA evidence with archaeological and historical contexts.
In our paper "Testing times: disentangling admixture histories in recent and complex demographies using ancient DNA," we addressed a fundamental question: how reliably can current methods such as qpAdm and the f3-statistic, detect and reconstruct admixture histories when genetic differences are small and migration histories complex? Through extensive simulations modeling both simple demographic scenarios and complex Eurasian population histories, we systematically evaluated the performance of these commonly used statistical approaches.
Our findings reveal that population differentiation is the primary factor determining qpAdm performance, with even complex gene flow histories having limited impact on overall accuracy. While these methods can typically identify the true ancestry model among a small set of candidates when analyzing historical-era populations, they often struggle to definitively reject all incorrect possibilities. Importantly, we developed practical guidelines to help researchers distinguish between competing models and better understand when their inferences are most reliable.
However, what happens when ancient settlement are connected through a "stepping-stone" model—analogous to Japanese garden pathways.
In collaboration with research led by the University of Ostrava Dr. Pavel Flegontov, we demonstrate that violations of qpAdm method assumptions which can be introduced by Stepping Stone Landscapes (SSL) – primarily due to complex, multi-directional gene flow patterns – can cause a poor correlation between statistical p-values and true model optimality. Within these demographic contexts, qpAdm inherently favors models with symmetrically arranged sources close to the target population, as these configurations consistently yield biologically plausible admixture fraction estimates (EAF between 0 and 1). Moreover, the fundamental violation of qpAdm's assumptions these demographic scenarios also allows a subset of highly asymmetric models, often involving distant sources, to coincidentally produce valid EAFs, causing them to appear plausible and contributing to false positive results. This can result in misinterpreting signals that might actually reflect isolation-by-distance effects as evidence for long-distance migration.
This research is published in our paper, "Performance of qpAdm-based screens for genetic admixture on admixture-graph-shaped histories and stepping-stone landscapes".
Page updated
Google Sites
Report abuse