Ansu Satpathy retweeted
I'm excited about this new class of generative models that allow us to understand how DNA sequence mediates regulation of transcription factor and nucleosome assemblies at single molecule resolution. 1/
1/4 One DNA sequence, many chromatin states.
Excited to share SMFNet, a diffusion model predicting single-molecule chromatin accessibility. Explainer video below!
With co-first author Till (his thread: bsky.app/profile/tschwa…), Yuliya, Roze, @anshulkundaje & Arnaud Krebs.
4
36
164
12,860
Ansu Satpathy retweeted
Last summer, I joined @intology, leading the effort to connect autonomous AI research to scientific discovery.
Today, we’re sharing work on autonomously adapting frontier scientific models to proprietary datasets and our broader vision for what's next.
🧵(1/6)
Today, we’re sharing work on autonomously adapting frontier scientific models to proprietary datasets and our broader vision for what's next.
🧵(1/6)
Locus sets the frontier for automating molecular design
Our automated research system Locus develops scientific AI models that outperform SOTA expert solutions
Unlocking the potential of scientific AI requires leveraging proprietary data. Locus achieves this autonomously 🧵👇
ss a number of tasks related to molecular scoring and design: training molecular property prediction models for a proprietary task, fine-tuning Boltz-2 to predict small-molecule potency, and building antibody design workflows around BoltzGen, Boltz-2, ESMC, and ESMFold2.
Each of these settings are real-world AI for science use cases, where tasks are often out-of-distribution for existing foundation models and limited in data, while featuring expensive and noisy evaluators. Compared to state-of-the-art baselines and LLM agents set to optimize these tasks, Locus achieves higher performance. Additionally, Locus generalizes better despite limited data, finds more sample efficient solutions, and explores a higher diversity of solutions.
7
11
83
5,808
Great work from @intology @ronusedh team! Locus can autonomously improve molecular design models across many use cases
Locus sets the frontier for automating molecular design
Our automated research system Locus develops scientific AI models that outperform SOTA expert solutions
Unlocking the potential of scientific AI requires leveraging proprietary data. Locus achieves this autonomously 🧵👇
ss a number of tasks related to molecular scoring and design: training molecular property prediction models for a proprietary task, fine-tuning Boltz-2 to predict small-molecule potency, and building antibody design workflows around BoltzGen, Boltz-2, ESMC, and ESMFold2.
Each of these settings are real-world AI for science use cases, where tasks are often out-of-distribution for existing foundation models and limited in data, while featuring expensive and noisy evaluators. Compared to state-of-the-art baselines and LLM agents set to optimize these tasks, Locus achieves higher performance. Additionally, Locus generalizes better despite limited data, finds more sample efficient solutions, and explores a higher diversity of solutions.
0
1
10
1,008
Ansu Satpathy retweeted
Locus sets the frontier for automating molecular design
Our automated research system Locus develops scientific AI models that outperform SOTA expert solutions
Unlocking the potential of scientific AI requires leveraging proprietary data. Locus achieves this autonomously 🧵👇
ss a number of tasks related to molecular scoring and design: training molecular property prediction models for a proprietary task, fine-tuning Boltz-2 to predict small-molecule potency, and building antibody design workflows around BoltzGen, Boltz-2, ESMC, and ESMFold2.
Each of these settings are real-world AI for science use cases, where tasks are often out-of-distribution for existing foundation models and limited in data, while featuring expensive and noisy evaluators. Compared to state-of-the-art baselines and LLM agents set to optimize these tasks, Locus achieves higher performance. Additionally, Locus generalizes better despite limited data, finds more sample efficient solutions, and explores a higher diversity of solutions.
Our automated research system Locus develops scientific AI models that outperform SOTA expert solutions
Unlocking the potential of scientific AI requires leveraging proprietary data. Locus achieves this autonomously 🧵👇
ss a number of tasks related to molecular scoring and design: training molecular property prediction models for a proprietary task, fine-tuning Boltz-2 to predict small-molecule potency, and building antibody design workflows around BoltzGen, Boltz-2, ESMC, and ESMFold2.
Each of these settings are real-world AI for science use cases, where tasks are often out-of-distribution for existing foundation models and limited in data, while featuring expensive and noisy evaluators. Compared to state-of-the-art baselines and LLM agents set to optimize these tasks, Locus achieves higher performance. Additionally, Locus generalizes better despite limited data, finds more sample efficient solutions, and explores a higher diversity of solutions.
30
39
165
53,208
I am honored to join the Board of Directors of 10x Genomics. For more than a decade, my lab has built on 10x technologies to study the immune system and cancer one cell at a time, and I’ve seen firsthand how they’ve changed what’s possible in biology and medicine. Single-cell and spatial tools have transformed science, and now generate the rich, high-resolution data that AI needs to learn the fundamental rules of biology. I look forward to working with Serge, Ben, Mike and the @10xGenomics team to advance AI-driven research across biology and biopharma, driving new discoveries and, ultimately, better diagnostics and therapies for patients.
We are pleased to announce the appointments of Ansuman Satpathy, MD, PhD, and Joseph Gilliam to our board of directors. Their collective expertise and leadership will be instrumental as we continue to empower scientific discovery. Learn more: 10xgen.com/27k
12
2
138
4,873
Ansu Satpathy retweeted
Another thing: AlphaFold is like a tiny, tiny, slice of all of biology. It's a great technical accomplishment, but it applies to static, well-formed structures (not intrinsically disordered proteins - roughly 30% of the entire proteome), and still cannot handle dynamic multimers well. It also depended on having enough data: 200,000 structures in the PDB and counting.
Saying that AlphaFold works, therefore we could "solve" biology with enough compute, is like saying that because we can predict the shape of every component in a car, we can predict everything the car will do on the road. And saying that compute or reasoning can somehow magically compensate for data is a religious belief.
Saying that AlphaFold works, therefore we could "solve" biology with enough compute, is like saying that because we can predict the shape of every component in a car, we can predict everything the car will do on the road. And saying that compute or reasoning can somehow magically compensate for data is a religious belief.
If you had an exponential amount of compute you could solve simulate biology exactly and wouldn't need to run any experiments.
Some of you will think "of course, but that's an unrealistic assumption" and others will think "no, you need experiments even with that much compute. That's just How It's Done".
You would of course need evaluations once you have a solution, but you don't in principle need experiments to find it: Experiments are a way to save compute, dating from a time when it was much more expensive.
And I think the presentation is exponential but real systems aren't irreducibly exponential. AlphaFold worked, after all.
16
26
279
9,853
Ansu Satpathy retweeted
@ClassiqTech has been acquired by @intel in a major step forward for the quantum ecosystem. Huge congrats to founders @MinerbiNir, @yehuda_naveh and @amir_naveh. And also to my friend @LipBuTan1 who is staking a leadership claim in the quantum world. @Wing_VC first invested in in 2020 and I've served on the Board ever since, a privilege I'm deeply grateful for
1
4
15
674
Ansu Satpathy retweeted
Congrats to Nir, Yehuda, Amir and the @ClassiqTech team on being acquired by Intel!
We’ve partnered with them since their 2020 Series A, and we couldn’t be prouder of the team. They’re among the finest founders we’ve ever had the privilege to back.
wing.vc/content/congra…
We’ve partnered with them since their 2020 Series A, and we couldn’t be prouder of the team. They’re among the finest founders we’ve ever had the privilege to back.
wing.vc/content/congra…
0
1
6
670
Ansu Satpathy retweeted
The Quiet Force. My profile of @AlexKolicich is out.
Raised by a single mom in Hamilton. Google engineer. Peter Thiel’s math guy. Founding partner of 8VC.
“8VC is 8VC because of Alex.” I was told.
I spent months reporting his story in SF, Palo Alto and Orange County, talking eople who knew him at Clarium, Mithril and 8VC.
Highlights
- Peter Thiel playing blindfolded chess
- Turning down Geoffrey Hinton’s lab twice
- Dreaming about oil
- Back story on 8VC's investments in Quince, Cognition and Anduril
Link to read below.
Raised by a single mom in Hamilton. Google engineer. Peter Thiel’s math guy. Founding partner of 8VC.
“8VC is 8VC because of Alex.” I was told.
I spent months reporting his story in SF, Palo Alto and Orange County, talking eople who knew him at Clarium, Mithril and 8VC.
Highlights
- Peter Thiel playing blindfolded chess
- Turning down Geoffrey Hinton’s lab twice
- Dreaming about oil
- Back story on 8VC's investments in Quince, Cognition and Anduril
Link to read below.
25
7
132
34,838
Ansu Satpathy retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com
890
2,645
19,911
4,700,610
Ansu Satpathy retweeted
Congrats to Link Cell Therapies on a big milestone! 🎉
First patients have been dosed with LNK001 at MD Anderson, the first clinical test of their logic-gated CAR-T platform.
Link also expanded its Series A to $90M and announced its leadership team:
✅ Jie D'Elia as CEO
✅ Dr. Friedrich Graf Finckenstein as CMO
✅ Co-founder Brian Slingerland as Executive Chair
Proud to have backed this team from day one and to keep supporting them as they take CAR-T into solid tumors.
prnewswire.com/news-releases/…
First patients have been dosed with LNK001 at MD Anderson, the first clinical test of their logic-gated CAR-T platform.
Link also expanded its Series A to $90M and announced its leadership team:
✅ Jie D'Elia as CEO
✅ Dr. Friedrich Graf Finckenstein as CMO
✅ Co-founder Brian Slingerland as Executive Chair
Proud to have backed this team from day one and to keep supporting them as they take CAR-T into solid tumors.
prnewswire.com/news-releases/…
0
1
7
608
Ansu Satpathy retweeted
I wish drug discovery were like pure math where, even if the problems can be fiendishly hard, they are bounded and closed with a well-defined solution and can be solved with pen and paper or with a computer. But unfortunately drug discovery (and biology more generally) is not n like physics, as I wrote about many years ago. A good reminder in the age of AI.
wavefunction.fieldofscience.com/2012/03/why-it…
wavefunction.fieldofscience.com/2012/03/why-it…
5
6
53
2,929
Cell–cell interactions are key to all biological processes, but there hasn't been a good way to perform inter-cellular screens for genetic regulators of these interactions in vivo.
We developed match-seq (multiplexed associative tagging of cell interaction histories) - using articles to transfer barcoded mRNAs from 'sender' to neighboring 'receiver' cells, so you can reconstruct cellular interactions with sequencing alone, no imaging required.
We coupled match-seq with CRISPR + scRNA-seq in cancer, revealing cancer-immune interaction neighborhoods at scale, and new cancer genes that can be targeted for tumor rejection via immune cell interactions.
Led by the brilliant @peterdu_ with Bassik lab!
biorxiv.org/content/10.648…
We developed match-seq (multiplexed associative tagging of cell interaction histories) - using articles to transfer barcoded mRNAs from 'sender' to neighboring 'receiver' cells, so you can reconstruct cellular interactions with sequencing alone, no imaging required.
We coupled match-seq with CRISPR + scRNA-seq in cancer, revealing cancer-immune interaction neighborhoods at scale, and new cancer genes that can be targeted for tumor rejection via immune cell interactions.
Led by the brilliant @peterdu_ with Bassik lab!
biorxiv.org/content/10.648…
CRISPR screens are great at telling you what a gene is doing inside your cells. But what if you're interested in what's happening to its neighbors?
Today, I'm excited to share match-seq, a new method that deciphers cell-cell interactions using barcoded RNA transfer.
1/9
8
77
405
44,065
Ansu Satpathy retweeted
1/3 Our Tahoe-100M paper is out in Cell today, in a special edition dedicated to modeling human cell. We released Tahoe-100M last year. It has been downloaded 600K+ times and it has become the basis for most serious efforts in building cell state models. We are proud of that. @tahoe_ai
13
47
205
34,517
Ansu Satpathy retweeted
Tahoe-100M became a seminal work. And we are proud that many are building on top. Read the full paper in today's special edition of Cell.
1/3 Our Tahoe-100M paper is out in Cell today, in a special edition dedicated to modeling human cell. We released Tahoe-100M last year. It has been downloaded 600K+ times and it has become the basis for most serious efforts in building cell state models. We are proud of that. @tahoe_ai
8
16
108
11,446
Ansu Satpathy retweeted
Four papers from the Wing community landed in the same issue of @CellPressNews this week.
The issue is focused on AI in biology, and it features:
- Tahoe-100M from @tahoe_ai's @nalidoust, @genophoria, @iamjohnnyyu, and others
- State from @genophoria and others
- scBaseCount from @genophoria and others
- New synthetic biology research on CAR T cells from @Satpathology's lab at Stanford
We can't think of another venture portfolio that has shown up this way in a single issue of Cell, and we're really proud of these teams.
Congrats to everyone behind this work!
Read the full issue here: cell.com/cell/issue?pii…
The issue is focused on AI in biology, and it features:
- Tahoe-100M from @tahoe_ai's @nalidoust, @genophoria, @iamjohnnyyu, and others
- State from @genophoria and others
- scBaseCount from @genophoria and others
- New synthetic biology research on CAR T cells from @Satpathology's lab at Stanford
We can't think of another venture portfolio that has shown up this way in a single issue of Cell, and we're really proud of these teams.
Congrats to everyone behind this work!
Read the full issue here: cell.com/cell/issue?pii…
0
2
10
690
Inflammation can leave a structural mark on the genome. Using base-pair resolution chromatin conformation data, @BenceDaniel2, Andy Chen, and @SandorDanielne show that immune memory to inflammation can be stored in the 3D genome. @PNASNews @4DNucleome
pnas.org/doi/10.1073/pn…
pnas.org/doi/10.1073/pn…
2
22
99
13,308
Ansu Satpathy retweeted
Thrilled to share that our work on genome-scale perturb-seq in primary human T cells is now published in Cell (cell.com/cell/fulltext/…) 🎉. The original post summarized most key results ⬇️ with a few key updates:
Together with Emma Dann, we are thrilled to present a massive new Perturb-seq atlas of 22M primary CD4+ T cells, from 4 donors, across 3 timepoints – the result of a decade-long collaboration between the Marson (@MarsonLab) and Pritchard (@jkpritch) labs.
🧵👇
3
14
111
13,081
Ansu Satpathy retweeted
TCR specificity prediction is advancing rapidly. The central question is not only which TCR sequence recognizes an antigen, but also which TCR pairings make recognition possible & where is the binding boundary? Our new paper tackles this question. 🧵
science.org/doi/10.1126/sc…
science.org/doi/10.1126/sc…
3
58
195
22,272
AI x fantasy football may be my perfect Venn diagram
Everyone is probably heading into the long weekend thinking about their fantasy football lineup. We know who is going to do well this season.
We just had our PhD engineers use Nori, our foundation model for predictions, to comb through 1,900 player-seasons and produce rankings 2026 fantasy football season.
We're giving it all away. Check out the full details on our blog (link in comments).
When we backtested our model against the 2025 season, Nori had 7.5% lower error than ESPN and 21.1% lower error than CBS, measured by mean absolute error against predicting total points.
The full top 200 rankings, methodology, and a reproducible notebook are all in
the blog post so you can run the model on your own league settings.
Good luck drafting.
0
0
7
1,125