Google DeepMind发布AlphaGenome

Understanding the genome is like understanding the language of life. What is the semantics of the genome of our DNA? How can we read it and how can we use it to understand ourselves?

I'm Dr. Gareth Hawks. I'm a lecturer from the University of Exeter. My research primarily focuses on how we understand what's called whole genome sequencing, which is where we try and measure every single base pair in the human genome. And I'm interested in understanding how changes in the genetic code might affect human phenotypes, so observable human traits. If you were to give every base pair 1 second of your time, you would spend tens of years reading it, basically.

I'm Sam Bryan. I'm a genomic analyst at the Center for Population Genomics. I find the genome incredibly fascinating by the fact that it has so many instructions spread out all across 3 billion base pairs of information, and and yet we understand so very little of it and how it interacts with each other.

The genome is the instruction manual for for cells. Alpha Genome [music] is a model that looks at chunk of the genome sequence and predicts what happens if you make a single mutation to this region.

And uh through kind of modeling the language of life, we understand some of [music] the genetic mutations, what they do, and whether they are harmful or not. The question now is how do you combine all these predictions into a single number.

The AVI score, the variant impact score, is a single number that tells you how deleterious or impactful this variant is. The higher that the number, the more likely it is to have [music] an effect on the phenotypes, such as cause disease. And we have precomputed this for all 9 billion possible single letter changes [music] in the human genome. Uh one of the core motivations for AVI was how do we go from 10,000 numbers that Alpha Genome [music] gives you to a single number that you can then use for efficiently prioritizing.

It scores every type of variant and tries to give you a AVI score for anything with a big impact.

The smaller we can make the haystack, the more likely we are to find the needle.

Alpha Genome Atlas is probably the most comprehensive and accurate [music] variant effect prediction platform. It's about 1 petabyte of data. Atlas [music] is both this massive data resource and it's also a website that we hope will make it [music] more accessible to biologists who are not proficient in code as well.

The team has built [music] the Genome browser, the website, where we could see which things light up in the genome from different perspectives and it was

[music]

to me like you're looking through a new new new window

at the starting to be a blueprint for how every base pair in the genome, no matter where it is, which chromosome it's on or which cell type it might be relevant to, how exactly that might affect human phenotypes.

With Alpha Genome Atlas, we are building up this thesaurus for the language of life.

I feel like any progress we make in this field can have enormous impact. AI is fundamentally changing how we do research and make scientific progress much faster than before.

We are working on this broader sort of agentic framework to combine all these different specialized models from AlphaFold to AlphaGenome to enable scientists to be able to solve the the biggest challenges that humanity is facing.