Google DeepMind Open-Sources AlphaGenome, a Foundation Model for Predicting Non-Coding DNA Function
DeepMind released model weights and a paper for AlphaGenome, which predicts the functional effects of non-coding genetic variants — opening the door to virtual screening of disease-linked mutations at scale.
Google DeepMind today released AlphaGenome, a foundation model designed to predict the functional impact of non-coding DNA — the vast, poorly understood stretches of the genome that don't encode proteins but play critical roles in gene regulation and disease. The paper and model weights dropped simultaneously, as @s6juncheng announced: "AlphaGenome paper and models are out today! We hope to open more use cases with the model weights!"
The significance is hard to overstate. Roughly 98% of the human genome is non-coding, and the majority of disease-associated genetic variants identified by genome-wide association studies (GWAS) sit in these regions. Until now, interpreting what those variants actually do has been one of genomics' most stubborn bottlenecks. AlphaGenome aims to solve this by enabling virtual testing — feeding a variant into the model and getting a prediction of its regulatory effect without running a wet-lab experiment.
Get our free daily newsletter
Get this article free — plus the lead story every day — delivered to your inbox.
Want every article and the full archive? Upgrade anytime.
No spam. Unsubscribe anytime.