I looked into the repo, and the expected cost per dataset is between $5 000-$10 000. This is a lot of money.
GPT-Rosalind surfaced a new lead for drug discovery: could restoring cholesterol transport protect neurons from degeneration?
DCTN1 is a gene linked to Perry syndrome, a rare neurodegenerative disorder. In mouse-brain data, disrupting it increased expression of genes involved in making cholesterol. GPT-Rosalind suggested a possible explanation: cholesterol may not be reaching where neurons need it.
That gives scientists a specific mechanism to test and, if confirmed, a direction for developing new treatments for neurodegenerative disorders.
With @perturbai_tx and @scrippsresearch, a team of thousands of GPT-Rosalind agents worked in parallel, connecting gene-expression results from millions of mouse brain cells with scientific literature in just 32 minutes. We’re now expanding this collaboration to identify new drug targets.
openai.com/index/scaling-ai-…
Florian retweeted
🚩 When Polymarket or Kalshi says there's a 70% chance something will happen, should we believe it?
Excited to share our new paper, just published in PNAS, in which we examined more than ONE MILLION forecasts from Polymarket, Kalshi, Manifold, and Metaculus, spanning 12 domains and forecasting horizons ranging from hours to years.
The main finding is striking: Prediction markets and collective forecasting platforms are remarkably well calibrated. When they assign events a 70% probability, those events really do happen about 70% of the time. And this pattern holds surprisingly well across different platforms, domains, and forecasting horizons.
But there's an important twist. When we looked more closely, we discovered systematic biases in some of the domains where forecasting matters most (wars, elections).
For decades, psychologists have documented how poorly humans often reason about probability and uncertainty. Our findings offer an encouraging counterpoint: When judgments and information are aggregated, we can achieve a surprisingly accurate understanding of an uncertain future.
That doesn't mean we can predict what will happen next with certainty. But it does mean that we can be remarkably good at estimating how uncertain the future really is.
This work was a great collaboration. The data for this project was meticulously collected for years by Justin Dickerson, the project was led by Hilla Shinitzky and Maayan Peer, with @mattansb leading the analyses.
📄 Read the paper:
pnas.org/doi/10.1073/pnas.26…
#Forecasting #Polymarket #Kalshi #DecisionScience #PNAS
"CEO Grace Zheng" Hmmm that rings a bell... Oh wait it's the lead author on the 10x Chromium debut paper 😄 Amazing!
8 million brain cells. 32 minutes. A new frontier in medicine.
In less than a year, PerturbAI built a proprietary AI-native drug discovery engine, created 6 genomic maps, revealing how genes influence health and disease in living systems.
Now, together with @OpenAI and @ScrippsResearch, we've uncovered 11,000 biological insights and ~100 promising leads using frontier AI in just 32 minutes.
We're expanding our partnership into drug-target discovery, bringing us closer to understanding disease and developing new medicines.
Better biology. Better targets. Better medicines.
Learn more: perturb.ai/news/perturbai-la…
Read the full blog here: openai.com/index/scaling-ai-…
I like this box actually. I hope we'll see them in clinics soon.
1/3 $TXG just announced Atera, its new imaging-based spatial transcriptomics platform, built on Xenium technology.
Can profile 18,000+ genes at subcellular resolution, detecting thousands of transcripts per cell.
Getting into scRNA-seq territory in terms of gene coverage...
Florian retweeted
The pace of scientific discovery is about to explode with how good tooling is getting.
This kind of analysis used to require a lab, a bioinformatics team, and all kinds of other resources.
I love tech that empowers users.
Heartbreaking to see. Apart from being a beautiful building, the Limoges town hall was also historically significant: members of Belgian government fled and sheltered there in 1940, after king Leopold III surrendered to the invading german army.
Florian retweeted
You Can See Everything has debuted with a 4.7 rating and is now the highest rated feature film on Letterboxd.
Florian retweeted
🧪🧭 What if we could ask a model which experiment gives us the cells we want?
Our new preprint, LabCompass, takes a step in this direction through generative inverse design. For me, this is one of the most exciting directions for the field.
Together with many others, we have been working on predicting how cells respond to perturbations. How can we use these models to design experiments that achieve a desired biological outcome?
🎛️ This inverse problem is difficult. Combinations and doses create spaces that cannot be exhaustively screened, and each experiment produces a distribution of cell states. Predictive models are also imperfect, particularly in unexplored regions where we want to discover something new.
LabCompass couples a generative model of experimental designs with a forward model of cellular responses. The desired phenotype guides generation in continuous design space, while a learned prior favors plausible designs. Selected candidates are tested experimentally, and the measurements inform the next round.
💡 The method is not tied to a particular virtual cell model. It can work with different differentiable forward models and cellular readouts, and the same trained models can address different phenotypic objectives.
🩸 We demonstrated this in primary human blood stem and progenitor cells. Across the study, we profiled 78 million cells and experimentally validated conditions enriching megakaryocytic, erythroid, monocytic and early lymphoid populations.
One result I particularly like: distinct culture recipes produced similar target enrichments. Several solutions to the same biological objective give us alternatives to investigate and develop further.
🔬 The opportunity extends beyond blood and cell culture. We also applied LabCompass to single-cell transcriptomic data from viral infection. I’m excited to explore its potential for other perturbation systems, drug combinations, differentiation and organoids.
This raises a question for virtual cell models: alongside prediction accuracy, how effectively can they guide experiments toward desired biological outcomes?
If this connects to a problem you are working on, I’d love to hear from you. We’re keen to see how LabCompass transfers to new systems and happy to explore ideas together.
A wonderful collaboration with Bertie Göttgens and Nicola Wilson at the Cambridge Stem Cell Institute. Huge credit to joint first authors Licyel Paulas-Condori, Alessandro Palma, Lorenzo Consoli and Juan Antonio Rubio-Lara, and the whole team across both labs!
📄 Preprint: biorxiv.org/content/10.64898…
💻 Code: github.com/theislab/LabCompa…
#InverseDesign #VirtualCell #SingleCell
Donating inference is going to be a popular mode of contribution in the next year
Validated and merged Rohan's PR. Big gain on a really difficult regime (that frankly I was stuck at). Incredible work.
κ = 2⁻¹⁵ (tightened from κ = 2⁻¹⁸²)
A 500 thousand fold improvement over the previous result and a 2 ^ 167 fold improvement over the original OpenAI result.
This result was a true community effort that combined techniques from a number of contributors. I want to take time to highlight each of them, but for now getting these results verified and published as fast as possible so everyone working on the problem is at the leading edge.
The data is also just too sparse and noisy. We will need 3-5x improvements in the chemistry of single-cell methods. Sampling more cells does not recover this issue
It would be cool if APIs could charge agents that call them with tokens, and those tokens could be re-sold on e.g. @OpenRouter
A Pearson correlation of 0.08 between predicted and observed perturbation effects - on datasets that cost millions to generate - should make us reconsider capital allocation in virtual cell efforts
The "Single-cell foundation models don't work for perturbation prediction" narrative was a measurement problem.
Our new paper is out in @NatureBiotech : "Deep Learning Perturbation Models Can Outperform Baselines on Calibrated Metrics"
Last year, several papers concluded that deep learning perturbation models couldn't beat simple baselines, a finding that cast doubt on the entire Virtual Cell direction.
We asked a different question: can the metrics used in those comparisons reliably tell a good prediction from a bad one?
Often, they can't.
Common metrics like MSE and control-referenced Pearson Δ frequently rank uninformative baselines above positive controls. With calibrated alternatives (eg, WMSE, weighted R²Δ, NIR), deep learning models do outperform mean, control, and linear baselines.
The data matters too. Norman19 is approximately 96% additive, and models see less than 1% of possible gene combinations during training. The ceiling for beating an additive baseline is nearly zero, and the training signal for non-additive interactions is almost absent. Some dataset rankings aren't model rankings, but just noise!
To address this, we developed the Dynamic Range Fraction (DRF): a measure of metric calibration. Before trusting a model ranking, verify that your benchmark can detect a useful prediction at all.
The models aren't shiny toys. The field just needed better rulers.
Paper: nature.com/articles/s41587-0…
Code: github.com/shiftbioscience/P…
Huge congratulations to the first authors @Henrymiller2012 and Gabriel Mejia, and the team of Shift Bio!
biology is not hypothesis generation bottlenecked. It’s data/verification bottlenecked
Biology world reaction to Anthropic finding is off base to me
Yes, no lab work was done — this is about analysis of biological data and generating hypotheses
And yes, this is routine stuff for a PhD or postdoc today
Well, when was AI doing equal level work in math? Somewhere between beating the Math Olympiad and solving Erdos problems. That was mid-to-late 2025
So non-lab biological thinking is ~9-15 months behind mathematical thinking.
Seems like a big deal!
“Its precise function or biotechnological utility (if any) is not yet clear”
you’ve discovered a hypothesis
Today we announced the Claude-led discovery of a molecular machine that we suspect could represent a new gene editing mechanism. Its precise function, biotechnological utility (if any), or level of significance is not yet clear, but at minimum it is work I would have been proud to do as a PhD student. The work was done mostly, though not entirely, by Claude: our life sciences team suggested a broad area of research, Claude read through the literature and a bunch of genome data and discovered something interesting, then Claude proposed experiments to verify the discovery and our team carried them out.
It’s easy to dismiss this as a one-off or curiosity, but we’ve repeatedly seen a pattern where AI performance in new intellectual domains goes from weak to superhuman in a matter of a few years. In 2023 models struggled to do math at the level of an average high-school student. In 2024 they started to do well on math competitions for the best high-schoolers in the country, in 2025 they started to solve minor open problems, in early 2026 more significant open problems, and in late 2026 they are beginning to solve the top few open problems in all of mathematics. We believe AI for biology is on a similar exponential trend.
The main difference between biology and mathematics, of course, is that math can be done purely theoretically, while biology requires experimentation. Some have used this to draw the conclusion that AI’s utility in biology will be limited. We think this is wrong. As we’ve demonstrated today, humans can collaborate with AI to perform the experiments, validate key results in a few weeks and, if necessary, work with the AI to iterate on what they find. Eventually it may even be possible for Claude itself to safely perform the experiments by autonomously controlling lab equipment, with appropriate safeguards in place, but we aren’t doing that today (our lab is also a BSL1/BSL2 facility that doesn't handle materials dangerous to humans).
More broadly, biomedical advancement has many stages — from fundamental biology discoveries, to translational research, to drug discovery, clinical trials, and finally the actual delivery of medicines and health care to patients. We are also interested in these later stages, but even simply accelerating the first stage of fundamental biological discoveries has the potential to speed up and broaden the entire pipeline. Improving our understanding of biology and sharpening biologists’ tools can drive forward all of the later stages, for example by identifying new drug targets, finding new therapeutic modalities, allowing for more precise measurement, and speeding up the experimental loop which itself further accelerates our understanding of biology. This will not in itself speed up clinical trial times, but if it succeeds it could greatly increase the number of promising candidates that go into the pipeline — an increase in throughput even though latency remains.
In Machines of Loving Grace, I wrote about AI’s potential to “cure most diseases in 5-10 years” — a goal that sounds impossible, but one I believe is just barely possible if AI is applied to every stage of the pipeline. The first step is showing that AI can first help with, and then drive, biological discoveries.
Claude’s discovery is the latest in a line of related prior work that goes back decades, beginning with systems like CRISPR, and continuing with discoveries like the bridge recombinase and VIPR in the past few years. Recently, there has been heightened interest in systems based on reverse transcriptase (RT) enzymes, the enzyme underlying the system Claude identified. And most recently, a Stanford team working independently described a novel RT system with an associated non-coding array that is in some ways similar to the one Claude found, though they are distinct systems that evolved independently from each other. I believe that we’re at the very beginning of finding such systems and developing them into powerful tools for biotechnology.
I’m proud of the resources Anthropic has invested in accelerating the public benefits of AI through the life sciences, and we’re aiming both to grow our life sciences team and to work with other scientists to extend this approach to a broad range of problems. If you have a proposal for a research collaboration or are interested in joining our life sciences team, please reach out.