AlphaGenome: DeepMind’s AI That Reads 1M Base Pairs And Designs Synthetic DNA Switches

Google DeepMind has unveiled AlphaGenome, an artificial intelligence that can analyse extremely long stretches of DNA—reportedly up to one million base pairs—and predict how subtle variants affect biological processes. The model is notable for its ability to interpret non-coding DNA, the roughly 98% of the genome once dismissed as “junk” but now known to regulate gene activity and influence disease.

AlphaGenome can help researchers identify mutations that cause disease and flag beneficial variants that may inspire new therapies. The system can also be used to design short stretches of synthetic non-coding DNA that do not occur in nature and could act as tissue-specific switches to control when and where a therapeutic gene is active.

How It Works

The model was trained on human and mouse genomes to learn how different sequences drive gene expression and other biological functions. By predicting the regulatory effects of long DNA sequences, AlphaGenome provides a higher-resolution view of the “dark genome” and how non-coding regions influence cells.

“AlphaGenome could be used to generate short stretches of non-coding DNA that don’t exist in nature,” said Ziga Avsec, who leads DeepMind’s genomics initiative. “In a gene therapy context, you would include this synthetic code in your delivery vector alongside the therapeutic gene. This ‘synthetic switch’ could help ensure that the therapy only turns on in the specific target tissue—like the retina or liver—while remaining silent in the rest of the body.”

Potential Applications

Applied safely, the technology could enable more precise, tissue-specific gene therapies, reduce off-target effects and accelerate discovery of new therapeutic targets across human, animal, plant and microbial genomes. Researchers also expect AlphaGenome to be a powerful tool for basic biology—helping to map regulatory logic, understand disease mechanisms and test synthetic regulatory sequences in silico before lab work.

Dr Robert Goldstone, head of genomics at the Francis Crick Institute, called the work a major milestone: “This level of resolution, particularly for non-coding DNA, is a breakthrough that moves the technology from theoretical interest to practical utility. One of the most remarkable demonstrations is its ability to predict gene expression from DNA sequence alone.”

Prof Rivka Isaacson of King’s College London described AlphaGenome as an exciting step toward illuminating the “dark genome” while cautioning that the full complexity of long non-coding sequences and biological feedback mechanisms means much work remains.

Caveats And Next Steps

Important technical and ethical challenges remain. Designing and delivering synthetic regulatory DNA safely requires robust experimental validation, careful assessment of off-target effects and regulatory oversight. AlphaGenome’s predictions will need extensive laboratory testing and clinical evaluation before any therapeutic application.

The research describing AlphaGenome was published in the journal Nature. The authors and external experts emphasize both the promise of the approach and the long path from computational predictions to safe, approved treatments.