On the limits of fitting complex models of population history to genetic data

Abstract

Our understanding of human population history in deep time has been assisted by fitting “admixture graphs” to data: models that specify the ordering of population splits and mixtures which is the only information needed to capture the patterns of allele frequency correlation among populations. Not needing to specify population size changes, split times, or whether admixture events were sudden or drawn out simplifies the space of models that need to be searched. However, the space of possible admixture graphs relating populations is vast and cannot be sampled fully, and thus most published studies have identified fitting admixture graphs through a manual process driven by prior hypotheses, leaving the vast majority of alternative models unexplored. Here, we develop a method for systematically searching the space of all admixture graphs that can incorporate non-genetic information in the form of topology constraints. We implement this findGraphs tool within a software package, ADMIXTOOLS 2, which is a reimplementation of the ADMIXTOOLS software with new features and large performance gains. We apply this methodology to identify alternative models to admixture graphs that played key roles in eight published studies and find that graphs modeling more than six populations and two or three admixture events are often not unique, with many alternative models fitting nominally or significantly better than the published one. Our results suggest that strong claims about population history from admixture graphs should only be made when all well-fitting and temporally plausible models share common topological features. Our re-evaluation of published data also provides insight into the population histories of humans, dogs, and horses, identifying features that are stable across the models we explored, as well as scenarios of populations relationships that differ in important ways from models that have been highlighted in the literature, that fit the allele frequency correlation data, and that are not obviously wrong.

Reconstructing the genetic history of Kra-Dai speakers from Thailand

“…A question arises: is it justified to put a lot of weight on a particular graph and derive historical interpretations of its topology if hundreds of diverse topologies fit the data equally well? This is a key point discussed by Maier et al 17 when revisiting admixture graphs from eight published studies. For instance, it was found that there are at least several models fitting the data significantly better than the admixture graph for East Asians used to support a key conclusion by Wang et al 13 , and the alternative models do not support the conclusion.…”

“…We note that the same constraints on the graph topology were applied as in the original study: French or Mbuti were assigned as an outgroup. Even a shallow exploration of both graph spaces found hundreds of models that fit the data as good as the published ones, and deeper exploration (performing more findGraphs runs and/or extracting more graphs from each run) is guaranteed to deliver further and further models of this kind 17 .…”

“…These conceptual problems are discussed in detail by Maier et al 17 , where novel approaches for admixture graph inference are introduced. Here we revisited the same population sets and graph complexities (the number of admixture events allowed) that were used by Kutanan et al 1 and explored these spaces of graph topologies with an automated tool, findGraphs 17 , that finds models corresponding to local fit optima in these spaces. In other words, we attempted to find alternative well-fitting models for the simple and complex graphs presented by Kutanan et al 1 , based on a very similar set of SNPs (HumanOrigins) and having the same population composition, the same complexity, and the same outgroup populations as in the original study.…”

“…As discussed in Maier et al 17 , inference of demographic history in the admixture graph framework has important limitations even if best-practice protocols introduced in that study are adhered to. For instance, it is unknown what parsimony level (the number of admixture events) is optimal for inferring true history, and changing graph complexity can dramatically change the pattern of topologies that fit the data, and hence their historical interpretation 17 . As discussed above, the outcomes of a model inference protocol also depend a lot on the choice of groups included in the model 17 .…”

Ancient DNA from Protohistoric Period Cambodia indicates that South Asians admixed with local populations as early as 1st-3rd centuries CE

“…We first performed pairwise qpWave modeling using the qpwave function from the R package “ ADMIXTOOLS2 ” 42 . We considered a pair composed of the target individual AB M-40/I1680 from Cambodia and one of the 13 reference populations as “left populations”.…”