### Post by Admin on May 14, 2016 22:43:46 GMT

Admixture Graph Modeling of high coverage ancient genomes

Strategy

We began with the Admixture Graph (Figure S6.1) in supplementary Information section 8, which relates the high coverage Mbuti (African outgroup), UstIshim, Malta1, Kostenki14, and GoyetQ116-1. This tree fits the allele frequency correlation patterns in the data to within the limits of our resolution, in the sense that the f2-, f3-, and f4-statistics among all possible pairs, triples, and quadruples of populations match the observed values within three standard errors (using an empirical standard error computed with a Block Jackknife).

We proceeded by attempting to fit representative high coverage (>1×) samples—Vestonice16, ElMiron, Villabruna, Loschbour—into the Admixture Graph in turn, starting with the oldest and moving forward in time. We evaluated whether each tested model was a fit to the data, again by testing whether the predicted values of all the f2-, f3-, and f4-statistics among all possible pairs, triples, and quadruples of populations matched the observed values, and assessing the significance of the difference using a Block Jackknife. In each of the admixture Graphs shown in this note, the labels on the solid edges give the estimated genetic drift in f2-units of squared frequency difference (parts per thousand). The labels on the dotted edges give mixture proportions.

Vestonice16 (~31,000 BP)

We added Vestonice16 to all possible nodes of Figure S6.1 either as a simple branch without mixture, or as a mixture between two branches. Altogether, we identified 3 models that fit the data (maximum |Z|<3). We show these in Figure S6.2:

Figure S6.2A shows a fit in which Vestonice16, Kostenki14 and GoyetQ116-1 have an unresolved splitting order: a trifurcation. Figure S6.2B shows Vestonice16 as a clade with Kostenki14, with GoyetQ116-1 as an outgroup to both of them. FigureS6.2CshowsVestonice16 mixed of lineages related to Kostenki14 and GoyetQ116-1.

ElMiron (~19,000 BP)

We added ElMiron to all possible nodes of the 3 models that fit the data for Vestonice16 (Figure S6.2). We added it either as a simple branch without mixture, or as a mixture of two branches. Figure S6.3 shows the three models that fit. Figure S6.3A shows ElMiron as a clade with GoyetQ116-1 and Vestonice16 as mixed. Figure S6.3B shows Vestonice16 as not mixed and ElMiron is mixed. Figure S6.3C shows both as mixed.

Villabruna (~14,000 BP)

We added Villabruna to all possible branches of the 3 models that fit the data for Vestonice16 and ElMiron (Figure S6.3), again either as a simple branch or as a mixture between two

branches. A total of nine models fit the data with at least three admixture events. We highlight two in Figure S6.4 (the other 7 models are qualitatively similar, with slight differences in the insertion points for the admixing lineages).

The models that fit the data for Villabruna have the shared features that:

- Vestonice16 is a mixture of lineages related to Kostenki14 and Villabruna.

- ElMiron is a mixture of lineages related to GoyetQ116-1 and a lineage that contributed most of the ancestry of Villabruna.

Strategy

We began with the Admixture Graph (Figure S6.1) in supplementary Information section 8, which relates the high coverage Mbuti (African outgroup), UstIshim, Malta1, Kostenki14, and GoyetQ116-1. This tree fits the allele frequency correlation patterns in the data to within the limits of our resolution, in the sense that the f2-, f3-, and f4-statistics among all possible pairs, triples, and quadruples of populations match the observed values within three standard errors (using an empirical standard error computed with a Block Jackknife).

We proceeded by attempting to fit representative high coverage (>1×) samples—Vestonice16, ElMiron, Villabruna, Loschbour—into the Admixture Graph in turn, starting with the oldest and moving forward in time. We evaluated whether each tested model was a fit to the data, again by testing whether the predicted values of all the f2-, f3-, and f4-statistics among all possible pairs, triples, and quadruples of populations matched the observed values, and assessing the significance of the difference using a Block Jackknife. In each of the admixture Graphs shown in this note, the labels on the solid edges give the estimated genetic drift in f2-units of squared frequency difference (parts per thousand). The labels on the dotted edges give mixture proportions.

Vestonice16 (~31,000 BP)

We added Vestonice16 to all possible nodes of Figure S6.1 either as a simple branch without mixture, or as a mixture between two branches. Altogether, we identified 3 models that fit the data (maximum |Z|<3). We show these in Figure S6.2:

Figure S6.2A shows a fit in which Vestonice16, Kostenki14 and GoyetQ116-1 have an unresolved splitting order: a trifurcation. Figure S6.2B shows Vestonice16 as a clade with Kostenki14, with GoyetQ116-1 as an outgroup to both of them. FigureS6.2CshowsVestonice16 mixed of lineages related to Kostenki14 and GoyetQ116-1.

ElMiron (~19,000 BP)

We added ElMiron to all possible nodes of the 3 models that fit the data for Vestonice16 (Figure S6.2). We added it either as a simple branch without mixture, or as a mixture of two branches. Figure S6.3 shows the three models that fit. Figure S6.3A shows ElMiron as a clade with GoyetQ116-1 and Vestonice16 as mixed. Figure S6.3B shows Vestonice16 as not mixed and ElMiron is mixed. Figure S6.3C shows both as mixed.

Villabruna (~14,000 BP)

We added Villabruna to all possible branches of the 3 models that fit the data for Vestonice16 and ElMiron (Figure S6.3), again either as a simple branch or as a mixture between two

branches. A total of nine models fit the data with at least three admixture events. We highlight two in Figure S6.4 (the other 7 models are qualitatively similar, with slight differences in the insertion points for the admixing lineages).

The models that fit the data for Villabruna have the shared features that:

- Vestonice16 is a mixture of lineages related to Kostenki14 and Villabruna.

- ElMiron is a mixture of lineages related to GoyetQ116-1 and a lineage that contributed most of the ancestry of Villabruna.