A genetic change can affect different cells in different ways, even though nearly every cell in the body carries the same DNA. Google DeepMind’s AlphaGenome 2 is designed to help researchers study those differences, offering predictions about how DNA variants may influence particular cell types and states.
Why the same DNA change can matter differently
Cells use genes in different patterns. Heart muscle cells, skin cells and immune cells, for example, switch different genes on and off. Those patterns can also shift with a cell’s state and activity.
That means a genetic variant does not necessarily have one uniform effect throughout the body. A change that matters in one type of cell may have a different impact in another, so broad analysis can miss effects relevant to a particular disease or treatment.
AlphaGenome 2 aims to make those distinctions easier to investigate. Rather than predicting effects only at a broad level, it is designed to estimate how genetic variants act in individual cell types and states.
Using cell data to sharpen genetic predictions
DeepMind trained the model on more than 100 million individual cell profiles across 68 tissues. Those profiles come from cell atlases: large repositories that catalog cells and their characteristics.
The company describes the approach as turning those static collections into a predictive tool for human genetics. According to DeepMind, AlphaGenome 2 identifies nearly twice as many variants that can change gene expression in cells as its previous state-of-the-art model.
The model can also make predictions for new cell types and biological states without being retrained, the researchers found. It may help identify where experimental data is missing, giving scientists clues about which questions to pursue in the lab.
From genetic signals to possible disease links
Adam Gayoso, a research scientist at Google DeepMind involved in the project, described the central advance as “tying human genetics to cell biology.” Researchers can start with variants found in large populations, then use the model to examine possible effects at the cellular level.
That finer view could help connect a genetic signal to the cells where it may be acting and, potentially, to disease. The predictions do not themselves establish that a variant causes illness or provide a treatment. They offer researchers a way to focus further investigation.
The cell-level detail also matters for drug research, according to Google DeepMind genomics lead Žiga Avsec. “When you develop drugs, you’re targeting specific proteins and specific cell types,” he told The Verge. A clearer picture of where a variant has an effect may therefore help scientists investigate targets with more precision.
More detail comes with practical tradeoffs
AlphaGenome 2 builds on AlphaGenome, but its more detailed predictions require more computing power and are “more expensive” to produce, Avsec said. That cost means it is not replacing its predecessor for every task.
In particular, DeepMind did not use AlphaGenome 2 to create its genome-wide Atlas covering “every possible DNA letter change in the human genome.” The source says the greater computational demands would have made that task impractical with the newer model.
Google is making AlphaGenome 2 available for non-commercial research through the AlphaGenome API, with a limited “private preview” for commercial research. That access gives researchers a chance to explore its predictions while the company continues work on the underlying models.
What could come next
DeepMind says it is generating more data, including for cell states that are not well represented today. Avsec said future versions could potentially draw on data “across multiple species,” which could give models a broader view of biology.
Over a longer horizon, models like AlphaGenome 2 could contribute to “virtual cells”: digital simulations intended to mimic biological systems. For now, the model’s role is more focused: use cell-level data to make genetic predictions more specific, and help scientists decide what to test next.