1.3 | An ABC framework for reconstructing complex
admixture histories
In this paper, we show how ABC can be successfully applied to
reconstruct, from genetic data, highly complex admixture histories
beyond models with a single or two pulses of admixture classically
explored with ML methods. To do so, we propose a novel forward-in-time
genetic data simulator and a set of parameter-generator and
summary-statistics calculation tools, embedded in an open source C
software package called MetHis . It performs independent SNPs or
microsatellites simulations under any two-source populations versions of
the Verdu and Rosenberg (2011) general model of admixture; and is
adapted to conduct ABC inferences with existing machine-learning ABC
tools implemented in R (R
Development Core Team, 2017).
We show that our MetHis -ABC framework can accurately distinguish
major classes of complex historical admixture models, involving multiple
admixture-pulses, recurring increasing or decreasing admixture over
time, or combination of these models, and provides conservative
posterior parameter inference under chosen models. Furthermore, we
introduce the quantiles and higher moments of the distribution of
admixture fractions in the admixed population as highly informative
summary-statistics for ABC model-choices and posterior-parameter
estimations.
We exemplify our approach by reconstructing the complex admixture
histories underlying observed genetic patterns separately for the
African American (ASW) and Barbadian (ACB) populations. Both populations
are known to be admixed populations of European and African descent in
the context of the Transatlantic Slave Trade, whose histories of
admixture remain largely unknown (e.g.
Baharian et al., 2016;
Martin et al., 2017). In this case-study,
we find that the ACB and ASW populations’ admixture histories are much
more complex than previously inferred, and further reveal the diversity
of histories undergone by these admixed populations during the TAST in
the Americas.