How can AI design genomes like Evo 2?

What Evo 2 does and why it matters

Evo 2 is a large biological foundation model trained on an enormous corpus of DNA — reported to be on the order of trillions of base pairs. By learning patterns across genomes from many species, the model can identify functional elements such as genes, regulatory sequences and splice sites, and generate short synthetic DNA sequences that follow biological design rules the model has learned.

Capabilities demonstrated so far

• Pattern recognition: Evo 2 can locate and annotate genetic features across diverse organisms.
• Sequence generation: it can propose novel DNA segments that resemble natural ones in statistical and functional terms.
• Cross-domain learning: because it was trained on data spanning many life forms, it can transfer insights from one clade to another, helping suggest candidates for gene design or regulatory tuning.

Potential uses

• Speeding early-stage research by prioritizing sequences for lab testing.
• Designing genetic parts for biotechnology, diagnostics, or crop and microbial engineering.
• Assisting basic research by revealing conserved motifs or regulatory logic across species.

Key limitations and safeguards

• Model outputs are hypotheses, not finished biological constructs: laboratory validation remains essential.
• Generative capacity does not equal the ability to create a living organism; building and controlling life requires complex cellular contexts and ethical oversight.
• The technology raises biosafety and biosecurity questions that call for responsible governance, transparency, and access controls.

In short, Evo 2 represents a powerful computational leap in how scientists can model and design DNA, turning vast sequence databases into actionable hypotheses. Its promise lies in accelerating discovery, but realizing that promise will require careful experimental follow-up and robust ethical frameworks.