Genome modelling and design across all domains of life with Evo 2

All of life encodes information with DNA. Although tools for genome sequencing, synthesis and editing have transformed biological research, we still lack sufficient understanding of the immense complexity encoded by genomes to predict the effects of many classes of genomic changes or to intelligently compose new biological systems. Artificial intelligence models that learn information from genomic sequences across diverse organisms have increasingly advanced prediction and design capabilities<sup>1,2</sup>. Here we introduce Evo 2, a biological foundation model trained on 9 trillion DNA base pairs from a highly curated genomic atlas spanning all domains of life to have a 1 million token context window with single-nucleotide resolution. Evo 2 learns to accurately predict the functional impacts of genetic variation-from noncoding pathogenic mutations to clinically significant BRCA1 variants-without task-specific fine-tuning. Mechanistic interpretability analyses reveal that Evo 2 learns representations associated with biological features, including exon-intron boundaries, transcription factor binding sites, protein structural elements and prophage genomic regions. The generative abilities of Evo 2 produce mitochondrial, prokaryotic and eukaryotic sequences at genome scale with greater naturalness and coherence than previous methods. Evo 2 also generates experimentally validated chromatin accessibility patterns when guided by predictive models<sup>3,4</sup> and inference-time search. We have made Evo 2 fully open, including model parameters, training code<sup>5</sup>, inference code and the OpenGenome2 dataset, to accelerate the exploration and design of biological complexity.

Authors

Institutions

Publication Details

Journal
Nature
Published
2026-03-04
DOI
https://doi.org/10.1038/s41586-026-10176-5
Citations
7
Primary Topic
Developmental Biology and Gene Regulation
Type
article
Field-Weighted Citation Impact
155.66
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Genome modelling and design across all domains of life with Evo 2

Daniel Chang, Jonathan C. Schmok, Michael T. Pearce, Brian Hie et al.
7 citations
Nature
Developmental Biology and Gene Regulation
155.66
article

Genome modelling and design across all domains of life with Evo 2

Daniel Chang, Jonathan C. Schmok, Michael T. Pearce, Brian Hie, Madelena Y. Ng, Mohsen Naghipourfar, Nicholas K. Wang, A. B. Vorontsov, Gwanggyu Sun, Nam‐Ky Nguyen, Aditi Merchant, Sudarshan Pinglay, Matthew G. Durrant, Chiara Ricci-Tam, Michael Poli, Alison Fanton, Ali Taghibakhshi, Garyk Brixi, Joseph Tey, Zachary Amador, Mingyu Liu, Mohammad R. K. Mofrad, Ken Janik, Hani Goodarzi, Jaspreet Pannu, Brandon Yang, Greg Zynda, Stephen A. Baccus, Kevin Zhu, Reshma Mehta, John St. John, Jerome Ku, Michael H. Herschl, Tina Hernandez-Boussard, Myra Deng, Thomas McGrath, Daniel Balsam, Elana Simon, Kimberly Powell, Ben Viggiano, Patrick D. Hsu, Patrick Collison, Eric Ho, Jeremy Sullivan, Samuel H. King, Anthony B. Costa, Euan A. Ashley, Daniel Guo, Etowah Adams, Rajesh Ilango, Liv Gorton, Dave P. Burke, David W. Romero, Gabriel A. Gonzalez, Amy X. Lu, David B. Li, Haoyu Dai, Greg Brockman, Steven Dillmann
article en
7 citations

Abstract

All of life encodes information with DNA. Although tools for genome sequencing, synthesis and editing have transformed biological research, we still lack sufficient understanding of the immense complexity encoded by genomes to predict the effects of many classes of genomic changes or to intelligently compose new biological systems. Artificial intelligence models that learn information from genomic sequences across diverse organisms have increasingly advanced prediction and design capabilities<sup>1,2</sup>. Here we introduce Evo 2, a biological foundation model trained on 9 trillion DNA base pairs from a highly curated genomic atlas spanning all domains of life to have a 1 million token context window with single-nucleotide resolution. Evo 2 learns to accurately predict the functional impacts of genetic variation-from noncoding pathogenic mutations to clinically significant BRCA1 variants-without task-specific fine-tuning. Mechanistic interpretability analyses reveal that Evo 2 learns representations associated with biological features, including exon-intron boundaries, transcription factor binding sites, protein structural elements and prophage genomic regions. The generative abilities of Evo 2 produce mitochondrial, prokaryotic and eukaryotic sequences at genome scale with greater naturalness and coherence than previous methods. Evo 2 also generates experimentally validated chromatin accessibility patterns when guided by predictive models<sup>3,4</sup> and inference-time search. We have made Evo 2 fully open, including model parameters, training code<sup>5</sup>, inference code and the OpenGenome2 dataset, to accelerate the exploration and design of biological complexity.

Nature
Johns Hopkins University (US), University of California, San Francisco (US), University of Washington (US), Stanford Health Care (US), BioQ Pharma (United States) (US), Film Independent (US), Palo Alto Institute (US), Nvidia (United States) (US), Emodo (United States) (US), Activated Research Company (United States) (US), OpenAI (United States) (US), Arc Research Institute, Columbia University (US), University of California, Berkeley (US), Stanford University (US), Core Laboratories (United States) (US)
Openalex Percentile: Top 0%
Developmental Biology and Gene Regulation
155.66
DOIOpen Access PDFOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.