←
Žiga Avsec465 posts
Žiga Avsec avatar

Žiga Avsec

@Avsecz

Genomics initiative lead at @GoogleDeepMind. Models from our team: Enformer, AlphaMissense, and AlphaGenome.

◷ Joined December 2015▧ 12 Photos and videos
Žiga Avsec avatar
Žiga Avsec@Avsecz·Nov 11

AlphaGenome 2 is live! A step change in performance and new capabilities over AG1! @adamgayoso's thread covers the key results. Manuscript: storage.googleapis.com/deepm… Blog: deepmind.google/blog/alphage… API: github.com/google-deepmind/a… A few notes under the hood & what's next:

Single-cell atlases continue to map the diversity of cell states, but how can we use them to interpret non-coding genetic variants? Excited to share AlphaGenome 2, our model for predicting cell-type-specific regulatory activity and variant effects from DNA sequence. deepmind.google/blog/alphage…

Peter Koo avatar
⟳ Žiga Avsec retweeted
Peter Koo@pkoo562·Nov 13

UPDATE on Genomics x AI blog! AlphaGenome 2 is now SOTA on a quantitative, zero-shot CRISPRi benchmark for seq2fun models. This is “unpublished” research shared fast with rapid engagement followed by corrections — essentially improving the work with real-time peer-review! 🧨💥

A benefit of our Genomics×AI blog is posts are living, versioned, updatable documents. So I've updated our enhancer benchmark with NTv3's revised release, Cerberus and AlphaGenome 2. Cerberus ties Borzoi overall but is worse on proximal and (maybe) better on distal enhancers 1/2

Adam Gayoso avatar
⟳ Žiga Avsec retweeted
Adam Gayoso@adamgayoso·Nov 12

We would like to learn about how people are using AlphaGenome 2! We'll spotlight selected projects so the community can learn from them, and some may be eligible for additional API quota or credits. Please submit by November 30th. docs.google.com/forms/d/e/1F…

Google DeepMind avatar
⟳ Žiga Avsec retweeted
Google DeepMind@GoogleDeepMind·Nov 12

Finding the exact genetic cause of a rare disease is like looking for a needle in a haystack. 🪡 On our podcast, @fryrsquared explores with @Avsecz and @adamgayoso how AlphaGenome 2 helps scientists test millions of DNA changes to find vital clues for human health. Timecodes: 00:00 Introduction to AlphaGenome 04:01 Genotype, Phenotype & DNA 08:30 Transformers to AlphaGenome 12:11 Mapping the Genome 17:45 Rare disease cases 21:10 AlphaGenome 2 30:06 Future vision 38:45 Hannah's thoughts

39:46
Anshul Kundaje avatar
⟳ Žiga Avsec retweeted
Anshul Kundaje@anshulkundaje·Nov 12

Very excited to see cell state conditioned sequence-to-omics models going mainstream. And as always @adamgayoso @Avsecz & the DeepMind genomics team have crushed it. 1/

Single-cell atlases continue to map the diversity of cell states, but how can we use them to interpret non-coding genetic variants? Excited to share AlphaGenome 2, our model for predicting cell-type-specific regulatory activity and variant effects from DNA sequence. deepmind.google/blog/alphage…

Demis Hassabis avatar
⟳ Žiga Avsec retweeted
Demis Hassabis@demishassabis·Nov 11

Super proud to announce AlphaGenome 2 - our new DNA sequence model! 🧬 It predicts the impact of DNA variants in different specific cell types - unlocking crucial information for understanding complex diseases. Available for academic research via our API: alphagenomedocs.com/

Your heart, brain, and immune system share the same DNA sequence. So how can a single-letter change affect disease in some cells, while sparing the rest? AlphaGenome 2 is our new AI model that zooms in on specific cells to help scientists understand what drives disease. 🧵

1:07
Pushmeet Kohli avatar
⟳ Žiga Avsec retweeted
Pushmeet Kohli@pushmeet·Nov 11

I’m very pleased to introduce AlphaGenome 2, @GoogleDeepMind’s next-generation genomics model that deciphers how DNA changes affect cells. Cells are the basic units of life and key to our understanding of biology. It is an ongoing challenge to predict how DNA changes affect the functioning of different cells, such as for a heart muscle cell or an immune cell fighting an infection, and to uncover cellular drivers of disease. To make progress on this challenge, we needed to design the architecture of AlphaGenome 2 so its predictions were conditional on cell types and states. This new design enables AlphaGenome 2 to predict variant effects across hundreds of human cell types and adapt to unseen cell and disease states without retraining. When evaluated on the TenK10K immune cell dataset, AlphaGenome 2 demonstrated a major leap in accuracy over its predecessor. It reliably identified nearly twice as many causal variants (30% vs. 16%) and directional gene effects (43% vs. 23%) compared to the original AlphaGenome. This step-change improvement in precision helps researchers pinpoint previously overlooked disease-relevant variants with high confidence. Using analysis on retrospective case studies, we were able to show AG2’s real world utility in disease understanding. A landmark study had previously shown that a variant in a distant regulatory switch drives irritable bowel disease by overactivating the ETS2 gene in gut macrophages. When used to make predictions across various immune cell types in the gut, AlphaGenome 2 singled out the exact causal variant and showed that it selectively boosts the ETS2 gene in gut macrophages, leaving other cell types and genes largely unaffected. I would like to thank my colleagues Adam Gayoso @adamgayoso, Žiga Avsec @Avsecz, and our entire genomics team for their incredible contributions and dedication to bringing this model to life. We are making AlphaGenome 2 available for non-commercial academic research via the AlphaGenome API, and soon we will release AlphaGenome 2 Skills in Google Antigravity. I’m incredibly excited to see how scientists around the world use this model to make progress on biological understanding and treating disease. Read more on our blog here: deepmind.google/blog/alphage…

Sundar Pichai avatar
⟳ Žiga Avsec retweeted
Sundar Pichai@sundarpichai·Nov 11

AlphaGenome 2 is helping uncover the cellular drivers of disease, giving scientists a powerful tool to predict how DNA variants impact specific cell types. Every cell in our body shares the same DNA blueprint, but different cell types (like heart muscle cells vs. immune cells) read that blueprint differently. This new @GoogleDeepMind model predicts how these specialized cells interpret DNA, turning genetic code into biological insights. Building on AlphaGenome 1 this makes it significantly easier for researchers to link genetic findings to the root causes of disease in the specific cells where they actually unfold.

Google DeepMind avatar
⟳ Žiga Avsec retweeted
Google DeepMind@GoogleDeepMind·Nov 11

Your heart, brain, and immune system share the same DNA sequence. So how can a single-letter change affect disease in some cells, while sparing the rest? AlphaGenome 2 is our new AI model that zooms in on specific cells to help scientists understand what drives disease. 🧵

1:07
Žiga Avsec avatar
Žiga Avsec@Avsecz·Nov 11

Extra non-commercial API quota request: forms.gle/ftBfHGH7NUrCZEjN6 Commercial API is rolling out on Google Cloud, currently in private preview. Model weights & AG2 Skills in Antigravity will be released in the future. Huge thanks to the @GoogleDeepMind team, early testers & data consortia!

Žiga Avsec avatar
Žiga Avsec@Avsecz·Nov 11

Important gaps remain open. Even after doubling sc-eQTL recall at high precision, AG2 misses the majority of high pip variants (e.g. outside open chromatin or post-transcriptionally). More data is needed: excited that we are working with Wellcome Sanger, Virtual Biology Initiative and others.

Žiga Avsec avatar
Žiga Avsec@Avsecz·Nov 11

AG2 also offers a modular foundation for future extensions: because scRNA-seq enters the frozen sequence trunk through a lightweight adapter, similar adapters could map spatial transcriptomics or experimental perturbation profiles into the same condition space.

Žiga Avsec avatar
Žiga Avsec@Avsecz·Nov 11

How do AG1, AG1 Atlas, and AG2 fit together? - AlphaGenome Atlas (generated with AG1) covers all 9B SNVs & stays fastest for large-scale rare-variant lookups. - AG2 supersedes AG1 for inference, adding cell-state resolution, primary-cell E2G/GWAS & zero-shot scRNA conditioning.

Žiga Avsec avatar
Žiga Avsec@Avsecz·Nov 11

In AG2, conditioning modulates representations across the trunk from the earliest layers (late conditioning barely helps!) and is trained across bulk assays & single cell RNA and ATAC atlases. It unlocks assay imputation and zero-shot prediction for new cell states.