The Big Leap to the Cellular Level
At its core, AlphaGenome 2 is an AI model designed to predict how tiny variations in our DNA affect the behaviour of specific human cells. Every cell in your body, from a heart muscle cell to an immune cell fighting infection, contains the same DNA blueprint.
The magic is in how each cell reads that blueprint, switching different genes on and off to perform its specialised job. Understanding this process is one of the biggest challenges in modern biology. Previous tools often looked at genetic effects across a whole tissue, which is like listening to an entire orchestra at once. AlphaGenome 2 can listen to a single instrument, predicting how a genetic variant plays out in one specific cell type, which could be the key to understanding a disease.
From AlphaGenome 1 to Version 2
The first AlphaGenome, introduced in 2025, was a significant breakthrough. It helped researchers understand how genetic mutations might affect biological processes by looking at their impact across broad tissues. However, this approach had limitations, as the effect of a crucial variant in a small number of cells could be diluted or lost in the tissue-wide average. DeepMind's team redesigned the architecture for AlphaGenome 2 to be 'conditional'. Instead of analysing everything at once, the new model takes the DNA sequence and a specific cell type as inputs. This focused approach allows it to learn the subtle rules that govern how different cells interpret the same genetic code, making its predictions far more precise and useful.
A Major Boost in Predictive Power
The results of this new approach are impressive. According to DeepMind, AlphaGenome 2 can identify nearly twice as many variants that alter gene expression compared to its predecessor. In a key test on a dataset of 28 immune cell types, the new model successfully identified the likely causal variant 30% of the time, compared to just 16% for the previous model. It was also far more accurate in predicting whether a variant would switch a gene on or off. This jump in performance is significant because it helps scientists filter through millions of genetic possibilities to pinpoint the specific changes that might drive disease, saving immense time and resources.
Pinpointing the Cellular Roots of Disease
So, what does this mean for our health? The applications are vast. Many common illnesses like inflammatory bowel disease (IBD), heart disease, and autoimmune disorders are linked to genetic variants that do their work in very specific cell types. For example, a recent study traced IBD risk to a variant that only affects gut immune cells. AlphaGenome 2 was able to computationally reproduce this discovery, identifying the exact variant and the cell it impacts. The model can also help researchers understand how our bodies react to things like vaccines by showing how variants behave differently in resting versus activated immune cells. By providing these highly specific insights, the model serves as a powerful new tool for drug discovery and the development of personalised medicine.
What Happens Next?
DeepMind has made AlphaGenome 2 available to non-commercial researchers through an API, with a private preview for commercial use. This follows the company's strategy with other major scientific models like AlphaFold, aiming to accelerate scientific discovery across the globe by making its tools accessible. The model is also designed to be adaptable; scientists can even upload data from their own experiments to see what the model predicts for new cell states it hasn't seen before. The ultimate goal is to move from simply reading the human genome to truly understanding its complex instruction manual, one cell at a time.
















