We’re launching AlphaGenome Atlas: an AI-powered searchable database mapping the predicted impact of all 9 billion possible single-letter DNA changes. Here’s how it could help researchers better understand our biology 🧵

AlphaGenome Atlas is over 30 times larger than the AlphaFold Database. It gives scientists an intuitive way to explore this vast 1-petabyte dataset - unifying interconnected resources so researchers can link genetic variants directly to the molecular mechanisms they disrupt.

Atlas includes the AlphaGenome Variant Impact (AVI) score, which combines AlphaGenome, AlphaMissense and other features to: 1️⃣ Rank mutations from low to high-impact 2️⃣ Reveal how they cause damage - like breaking gene switches or RNA splicing instructions.

We believe foundational biology tools should be accessible to everyone. AlphaGenome Atlas resources are available to the global scientific community via the Atlas website, our AlphaGenome API, as a skill in Google Antigravity, and coming to @GoogleCloud soon →

At this rate I wouldn't be surprised if AlphaFold cures cancer before Gemini gets good

mapping 9B DNA variants in a browser is nice until u realize all that compute is locked inside google closed stack... drop the open weights

9 billion possible DNA changes is far too much for a person to study one by one. This is where AI can become incredibly useful. It can search through enormous amounts of biological information and help scientists find the changes worth investigating. Some of AI’s biggest breakthroughs may happen in science, not chatbots.

predictions have more scientific value when labs can query them without depending on one cloud

genuinely mindblowing , genome sequencing at the light of speed now!

Now we need every datacenter in the world working to solve human disease and discover how we can radically extend human lifespan.

9 billion DNA letter-swaps… and now there’s a search box. Wild that biology got a query language before half our agent desks did.

Humans after 10,000 years: “Maybe we should spellcheck the source code.”

AlphaGenome predicts what a DNA letter change does. Prime editing is the molecular word processor that can find-and-replace that letter. Alphabet Inc. holds ~16.56 million shares of Prime Medicine $PRME, roughly 9.2% of the company

Here is a free app to interact with AlphaGenome.

We've spent decades building tools to read biology. Now we're building AI systems that can help us interpret it at enormous scale. Mapping the predicted impact of 9 billion possible DNA changes is a pretty incredible demonstration of where biology and AI are heading.

Unfortunate timing for breaking the news. Whole X busy with Navier Stokes drama. Great tho!

Now do epistasis.

how many parameters in alpha genome?

A Common Thread

This is the type of work that will improve publics perception of AI. This is why we need to 3x the amount of data centers we have in the U.S.

It's like building a spellchecker for the entire human blueprint. Now we've just got to hope the "autocorrect" doesn't start making its own creative decisions.

What interests me most is the shift from sequencing the genome to navigating it. Once billions of variants become searchable, the bottleneck moves downstream: which predictions deserve experiments, and how efficiently can we get reality to answer? That may be the bigger story behind AlphaGenome. DeepMind Just Turned the Genome Into a Giant Google Map

9b nice, can finally become an at home biologist 😀

A searchable atlas at that scale is wild. How are you exposing model confidence and uncertainty so researchers can separate useful signal from a precise-looking guess?

I am more interested in understanding of ai can do some research like scientist, even if it's a small one

this is where AI in science gets really exciting precomputing the effects at this scale could save researchers a massive amount of time

Great use of narrow AI! This will help to better understand human biology. Another Nobel on the horizon?

dna but with find :D

9 billion single-letter changes is the sort of number that needs a search box

Now THIS is what AI should be for!

So expecting gemini 4 to be a beast scientist

@sundarpichai Single-letter DNA Changes are scientifically / genetically called "Point Mutations"

Free for researchers is huge too

Good to see something that can actually help people understand and research biology instead of locking it behind classifiers like some labs do

Please do Arabidopsis next 🤌🏻

This almost certainly opens up lots of old research projects that were shut down because they hit a wall. It’s kind of like what happened with neural networks. Lots of promising research, but it hit a wall, and we entered a cold stretch. Then AlexNet comes along, and reignites tons of old research that could only be done once new tech had unlocked the necessary capabilities. So it’s not even the new research (though obviously I’m excited about that too). Now that this has been solved, teams can go back and continue research in places they had already made lots of progress (but didn’t complete it).

Mapping billions of biological variations requires absolute data integrity. When you handle scale like this, automated pipelines still need strict verification gates and human oversight to prevent silent errors.