View Full Version : A third archaic introgression into Asia and Oceania?

17-01-19, 21:06

Mayuk Mondal et al:
"Approximate Bayesian computation with deep learning supports a third archaic introgression in Asia and Oceania"

"Since anatomically modern humans dispersed Out of Africa, the evolutionary history of Eurasian populations has been marked by introgressions from presently extinct hominins. Some of these introgressions have been identified using sequenced ancient genomes (Neanderthal and Denisova). Other introgressions have been proposed for still unidentified groups using the genetic diversity present in current human populations. We built a demographic model based on deep learning in an Approximate Bayesian Computation framework to infer the evolutionary history of Eurasian populations including past introgression events in Out of Africa populations fitting the current genetic evidence. In addition to the reported Neanderthal and Denisovan introgressions, our results support a third introgression in all Asian and Oceanian populations from an archaic population. This population is either related to the Neanderthal-Denisova clade or diverged early from the Denisova lineage. We propose the use of deep learning methods for clarifying situations with high complexity in evolutionary genomics."

18-01-19, 00:17
Fig. 2
Demographic models implemented for explaining the genetic variation present in current African (A), European (E), East Asian (Ea), Andamanese (An), Indian (I), Papuan (P) and Australian (Au) populations, in relation with Altai Neanderthal (N) and Denisovan (D) archaic populations.

The ghost archaic population is most likely to be a Denisovan-Neanderthal admixed population (Xe) and Model H is the best fit for their hypothesis. East Asians' additional Neanderthal ancestry compared to Europeans can be explained by gene flow from this extinct hybrid population (2.6%) at 51 Kya. Denisova 11 from Denisova Cave was 19.6% Neanderthal and genetic admixture between Neanderthals and Denisovans was more common than we thought.

We first evaluated the performance of the proposed ABC-DL methodology for distinguishing between the proposed eight models. We used simulated data as pseudo-observed data and run the full ABC-DL pipeline; we applied a ‘hard’ model classification to assign the simulated data to the model with the highest posterior probability (see Supplementary Note 2). The confusion matrix (Supplementary Table 6) summarizing the model classification shows that the proposed ABC approach identifies the model that generated the data (Modelsim) with posterior probability P(Modelsim = Modelabc | Data) > 50% in all the models. When comparing the eight considered models by ABC-DL with the observed data, the ABC model inference supports model H (P(Model = H|Data) = 0.46), namely the introgression of genes from an archaic Denisovan-Neanderthal admixed extinct population (Xe) into Asian AMH populations, before the split of Oceanian groups, closely followed by Model F (P(Model = F|Data) = 0.38), which models Xe as a sister population of Denisovans. Model H is slightly more likely than Model F (1.2 times more likely), and much more likely than models E (5 times more likely), G (7.6 times) and D (107 times). Models not considering the presence of an extinct archaic ‘ghost’ population introgressing AMH -A, B, C and D- obtained a posterior probability of ~0 (see Supplementary Figure 4). Therefore, only models with admixture of an extinct population in the Neanderthal–Denisovan clade, but with significant differences to each of them, are supported by the ABC-DL framework.