A variational autoencoder (VAE) is used to learn a reduced representation of the SNP dataset by minimizing the difference between the input SNPs and the reconstructed SNPs.
The mean (μ) and standard deviation (σ) are used for plotting. The black circles indicate the μ. The transparent colors show the σ (100 sampled "Z" or latent variables for each organism). For the following animation, we show the change of μ and σ during optimization (1 frame per 10 epoches). Each color represents a different species.
For more info, see pre-print here: https://www.biorxiv.org/content/early/2018/09/28/429662