Can AI write genomes?

Large genome models are changing how biology is designed

Researchers have trained large AI models on massive DNA sequence datasets so the systems can learn patterns that appear across genomes. One such foundation model was trained on trillions of DNA bases and is reported to generate short genome sequences, identify regulatory elements, and predict molecular features across a wide range of organisms. These capabilities mark a step toward AI-guided genome design that could speed research in basic biology, biotechnology and medicine.

Potential scientific and practical benefits

• Faster discovery: AI can prioritize candidate genes, promoters or synthetic constructs for laboratory testing, reducing trial-and-error in the wet lab.
• Design of new functions: models may suggest genetic changes to improve protein production, metabolic pathways, or the stability of engineered organisms used in manufacturing.
• Cross-domain learning: because the models are trained on diverse life forms, they can transfer insights between microbes, plants and animals, aiding comparative genomics and translational research.

Risks, limits and governance

• Model outputs require experimental validation; computational suggestions are not equivalent to functional proof in living cells.
• The ability to design genetic sequences raises biosafety and biosecurity concerns: misuse or accidental release of harmful constructs is a real hazard.
• Ethical and regulatory frameworks lag behind the technology; careful oversight, access controls, and community standards are needed before widespread deployment.

Next steps for the field include tight integration of AI predictions with laboratory pipelines, clearer standards for validation, and international discussion on responsible use. If managed carefully, genome-scale AI tools could accelerate beneficial applications while demanding stronger safety and governance measures.