EVO: DNA FOUNDATION MODELING FROM MOLECULAR TO GENOME SCALE

Summary from the TLDR newsletter

Evo is a long-context biological foundation model capable of both prediction tasks and generative design, from molecular to whole genome scale. The model generalizes across the fundamental languages of biology, DNA, RNA, and proteins to generate DNA sequences using a context length of 131k tokens. It is based on StripedHyena, a deep signal processing architecture designed to improve efficiency and quality over prevailing Transformer architecture. A playground is available where users can generate DNA in their browsers using Evo.