 Abstract variational auto-encoders, VAEs, have been used to analyze protein data, such as globins, lactamases, ion channels, and transcription factors. These VAEs can cluster sequences into groups based on their phylogeny, and also generate new sequences that retain the statistical properties of protein composition. Additionally, these VAEs can be used to study the latent generative landscape of proteins, which can capture phylogenic groupings, functional and fitness properties, and provide insight into directed and natural protein evolution. By combining generative properties and functional predictive power of VAEs and co-evolutionary analysis, it may be possible to use them in applications for protein engineering and design. This article was authored by Cheyenne Ziegler, Jonathan Martin, Claude Sinner, and others.