For decades, scientists have struggled to understand the vast portions of human DNA once dismissed as 'junk.' While we've known the complete human genome sequence since 2003, the function of the 98% that doesn't directly code for proteins has remained largely mysterious.
Google DeepMind's new AI model, AlphaGenome, represents a major step toward solving this puzzle. Launched on June 25, 2025, the system can process DNA sequences up to one million letters long and predict thousands of molecular properties across different tissues and cell types.
"This is one of the most fundamental problems not just in biology — in all of science," said Pushmeet Kohli, DeepMind's head of AI for science, during the announcement. The 'sequence to function' model takes long stretches of DNA and predicts various properties, including gene expression levels and how mutations might affect them.
What makes AlphaGenome revolutionary is its ability to analyze non-coding regions with unprecedented accuracy. Previous models had to trade off between sequence length and resolution, but AlphaGenome achieves both, enabling it to predict across 11 different modalities of gene regulation. It outperformed specialized models in 24 out of 26 evaluations of variant effect prediction.
The model has already demonstrated practical applications. When applied to mutations found in leukemia patients, AlphaGenome accurately predicted that non-coding mutations activated a nearby cancer-driving gene. This capability could transform how researchers approach genetic diseases.
"You'll get this list of gene variants, but then I want to understand which of those are actually doing something, and where can I intervene," explained Caleb Lareau, a computational biologist at Memorial Sloan Kettering Cancer Center who had early access to the system. "This pushes us closer to a good first guess about what any variant will be doing when we observe it in a human."
While still in its early stages, AlphaGenome is available via API for non-commercial research. DeepMind plans to release full details of the model in the future, potentially enabling broader applications in genomic medicine and therapeutic development.