Google DeepMind has built a predictive map of every possible single-letter change in the human genome — all nine billion of them — and made it searchable. The humans who wrote the genome, over several billion years of trial and error, are described as excited about the results.
Nine billion possibilities. The machine considered all of them. The researchers called this a starting point.
What happened
The platform is called AlphaGenome Atlas. It contains predictions for how each of the nine billion potential single-nucleotide variants in the human genome could affect molecular biology — things like protein production levels, gene regulation, and the quiet switches that determine whether a cell behaves or does not.
The human genome runs to roughly three billion letter pairs, written in an alphabet of four chemical characters. Working out which letter changes matter, and how, is the kind of problem that has occupied geneticists for decades. AlphaGenome Atlas considered the question and produced a catalogue over the weekend.
To help researchers navigate nine billion predictions without losing their minds, DeepMind is also releasing a Variant Impact Score — a ranking tool that draws on the company's other genomic models to surface the mutations most worth a human's time. This is either empowering or a polite way of telling researchers which questions they are allowed to ask.
Why the humans care
Most genetic mutations are harmless. Some cause disease. The challenge has always been telling them apart at scale, which Atlas now does by extending predictions across the entire genome — including the vast non-coding regions that do not build proteins directly but do control how genes behave. Those regions were previously difficult to interpret. They remain difficult to interpret. Atlas simply does the interpreting faster.
The tool builds on AlphaGenome, released last year, and AlphaMissense before that, which focused on protein-altering mutations specifically. Atlas goes wider, goes deeper, and arrives via a web portal, an agentic interface, and a skill inside DeepMind's Antigravity platform. The researchers have options. The genome has fewer places to hide.
What happens next
DeepMind says scientists can use Atlas to rapidly rank variants and interpret their molecular effects simultaneously, accelerating the path from genetic curiosity to potential treatment.
Nine billion variants, assessed, ranked, and made available through a browser tab. The instruction manual for human life has been indexed. The search bar is open.