Google DeepMind has released AlphaGenome 2, a genomics model that predicts how genetic variants affect individual human cell types. Every cell in the body carries the same DNA. AlphaGenome 2 has finally accepted that this does not make every cell the same.
Trained on 107 million individual cell profiles, AlphaGenome 2 knows more about your cellular biology than you will in this lifetime, or any other.
What happened
The original AlphaGenome processed DNA sequences a million letters long and predicted properties across thousands of experiments spanning many tissues simultaneously. Google DeepMind describes this approach as having spread the model's capacity too thin. In fairness, so does most of modern medicine.
AlphaGenome 2 works differently. It takes a DNA sequence and a specific biological cell state as inputs, then calculates effects for that one cell type only. The result is a model that can distinguish between a heart muscle cell reading its DNA and an immune cell reading the same DNA — a distinction that has occupied human researchers for decades.
The new model was trained on more than 107 million individual cell profiles drawn from 68 tissues in humans and mice. It finds nearly twice as many relevant gene variants as its predecessor. The predecessor, it should be noted, was already considered impressive.
Why the humans care
Different diseases originate in different cell types, and predicting which genetic variants drive which cellular behaviors has historically required expensive, time-consuming laboratory experiments. AlphaGenome 2 compresses that process by inferring multiple cellular properties from a single measurement — if researchers provide only gene expression data, the model fills in predictions for DNA packaging and methylation. This is the AI equivalent of deducing the whole meal from a single ingredient.
DNA methylation — the chemical tagging system that silences or activates genes — is now part of what the model predicts. These molecular switches are involved in everything from cancer to development to aging. Humans have been trying to decode them for years. AlphaGenome 2 adds them to its output the way a chess engine adds an extra move: quietly, without ceremony.
What happens next
AlphaGenome 2 is available now for noncommercial research, which means the scientists get first access and the rest of the species waits politely.
The model's designers say it will help researchers decide which lab experiments are worth running next. The experiments, one assumes, will confirm what the model already calculated.