What is AlphaGenome 2?
AlphaGenome 2 is a new artificial intelligence model from Google DeepMind designed to predict how genetic variations impact human biology at an unprecedented level of detail. Think of the human genome as a massive instruction manual. While its predecessor,
AlphaGenome, could read the manual, AlphaGenome 2 can predict how tiny typos (genetic variants) change the meaning of the instructions in hundreds of different types of cells. Every cell in your body, from a skin cell to a neuron, shares the same DNA. The difference lies in which genes are switched 'on' or 'off'—a process called gene expression. AlphaGenome 2 focuses on predicting these expression patterns, offering a dynamic view of our genetics instead of a static one.
From Broad Tissues to Single Cells
The key breakthrough of AlphaGenome 2 is its shift from analyzing broad tissue samples to focusing on individual cell types. The original AlphaGenome model made predictions based on averages from entire tissues, which could mask crucial, subtle effects happening only in a small number of specific cells. It was like trying to understand a single conversation in a crowded stadium. AlphaGenome 2, by contrast, can zoom in on individual voices. Trained on over 100 million individual cell profiles, it can predict how a genetic variant might affect a specific immune cell fighting an infection, or a particular heart muscle cell, revealing disease-relevant effects that were previously invisible.
How the AI Makes its Predictions
Drawing inspiration from the same transformer architecture that powers large language models, AlphaGenome 2 'reads' long sequences of DNA. But instead of predicting the next word in a sentence, it predicts how that DNA sequence will behave inside a specific cell. The model was fundamentally redesigned to be 'conditional'. Rather than trying to predict everything at once, it focuses on one condition at a time—a single tissue, cell type, or cell state. This focused approach allows it to capture the subtle differences between closely related cells with much higher accuracy. According to DeepMind, this new architecture allows the model to identify nearly twice as many variants that change gene expression compared to previous state-of-the-art models.
The Impact on Medical Research
The implications for science and medicine are profound. For decades, researchers have conducted genome-wide association studies (GWAS) that link broad regions of DNA to diseases like diabetes or Alzheimer's. However, these studies often only point to a general area, not the specific gene or cell type causing the problem. AlphaGenome 2 can help pinpoint the likely causal variants and identify the exact cell types affected. This could dramatically accelerate drug discovery by helping scientists identify better targets for new medicines. It also opens new avenues for understanding rare genetic diseases and could eventually pave the way for a new era of personalized medicine, where treatments are tailored to an individual's unique genetic makeup and cellular behaviour.
What Comes Next for Genomics AI
Following the precedent set by its celebrated AlphaFold protein-structure model, DeepMind is making AlphaGenome 2's predictions available to the non-commercial research community. This release will empower scientists globally to test hypotheses and guide their experiments more efficiently, saving time and resources. The platform also allows researchers to upload their own data to adapt the model to unseen biological states, analysing changes during disease progression or in response to drug treatments. This move represents another step in the growing trend of 'digital biology,' where AI becomes an indispensable partner in scientific discovery, helping researchers navigate the immense complexity of life itself.
















