Visualizing population structure with variational autoencoders. Issue 1 (18th January 2021)

• Record Type: Journal Article
• Authors: Battey, C J; Coffing, Gabrielle C; Kern, Andrew D
• Editors: Sethuraman, A
• Is Part Of: G3. Volume 11:Issue 1(2021)
• Publication Date: 2021-01-18
• Subjects: population structure -- population genetics -- data visualization -- pca -- variational autoencoder -- deep learning -- machine learning -- neural network; Genetics -- Research -- Periodicals; Genomics -- Periodicals; Genetics; Genomics; Genes; Genetics -- Research; Genomics; Electronic journals; Periodical; Periodicals; Fulltext; Internet Resources; Periodicals; 572.8
• DOI: 10.1093/g3journal/jkaa036
• Languages: English
• ISSNs: 2160-1836

Abstract

Abstract: Dimensionality reduction is a common tool for visualization and inference of population structure from genotypes, but popular methods either return too many dimensions for easy plotting (PCA) or fail to preserve global geometry (t-SNE and UMAP). Here we explore the utility of variational autoencoders (VAEs)—generative machine learning models in which a pair of neural networks seek to first compress and then recreate the input data—for visualizing population genetic variation. VAEs incorporate nonlinear relationships, allow users to define the dimensionality of the latent space, and in our tests preserve global geometry better than t-SNE and UMAP. Our implementation, which we call popvae, is available as a command-line python program at github.com/kr-colab/popvae. The approach yields latent embeddings that capture subtle aspects of population structure in humans and Anopheles mosquitoes, and can generate artificial genotypes characteristic of a given sample or population.