What can Evo 2 do with DNA?

Question

Hans Steiner · Accepted Answer

An AI foundation model trained on genomes can predict and generate sequences

Researchers have developed a large biological language model trained on an enormous corpus of DNA—trillions of base pairs spanning many species. The system can identify features in genetic sequences such as genes, regulatory elements and splice sites, and it can propose new short DNA sequences that resemble functional motifs. Because it learns statistical patterns across the tree of life, the model can generalize patterns that are difficult to capture with traditional rule‑based methods.

Potential scientific uses

Rapid annotation: accelerate the identification of genes and regulatory regions in newly sequenced organisms.
Design assistance: suggest candidate sequences for synthetic biology tasks, such as promoters, genetic parts or small constructs for laboratory testing.
Hypothesis generation: reveal conserved sequence features and propose experiments to probe function.

Risks and safeguards

Experimental validation remains essential: AI‑generated sequences are predictions that must be tested in the lab; biological systems often behave differently than models expect.
Dual‑use concerns: the same generative abilities could be misused to produce harmful biological agents, so responsible release and governance are necessary.
Ethical and regulatory framing: developers, funders and regulators must work together to set standards for testing, transparency and limits on capabilities accessible without oversight.

This technology is a powerful research tool that could speed discovery and engineering in biology, but realizing benefits while minimising harm will require strict validation pipelines, biosafety review and international coordination.