Estimate of Ancestry Coefficients and Population Structure
Sparse non-negative matrix factorization (sNMF) from the R package LEA v1.4.0 (Frichot, Mathieu, Trouillon, Bouchard, & François, 2014) was used to reconstruct ancestry coefficients among sampled individuals in the fully-filtered dataset . To evaluate the most likely number of ancestral populations (K), we computed the cross-entropy criterion for K=1 to K=20 with ten replicates each. Since the lowest cross-entropy was not clear, we present multiple levels of K to interpret the ancestry coefficients (see results). The summed ancestry coefficients for each population were plotted in geographical pie charts to visualize geographical population structure (Figure S1 ).