Sabitlenmiş Tweet
Matthew Osman
347 posts
Matthew Osman
@thematthewosman
https://t.co/pgViLUSkmg
Katılım Şubat 2025
542 Takip Edilen576 Takipçiler
Delighted to say that I started at @ScienceCorp_ earlier this month to build biohybrid BCIs. After speaking with @mardinly and @maxhodak_ I knew immediately that this is what I wanted to work on next.
English
This is clearly what is going to be needed for AI to make in-roads into biology
Fabio Boniolo@FabioZB_I
We started building Polyphron to manufacture tissues that can replace lost function at scale. What became clear along the way is that if our tissues are good enough to implant, they're also the ideal substrate to close the sim-to-real gap, giving AI models of biology a physical test unit to check their predictions against human ground truth. Each tissue unit thus becomes an oracle. You dose it, perturb it, and record how it responds over hours and days, creating a continuous, multimodal movie of what actually happens when you intervene on living human tissue.
English
Matthew Osman retweetledi
what does a neolab that finds PMF with scientists and biotech looks like in terms of scale of revenue and industry utility. this is the commercial question for bioML.
Forming the best bioML teams works for mid-size acquisitions.
> @EvoscaleAI acquired by @czi
> @CoefficientBio acquired by @AnthropicAI
Or mid 8-figure deals
> @chaidiscovery
> (kinda @NOETIK_ai though they have a true data moat)
But excellent PMF with models is where we get truly massive outcomes.
Excited for this
Dr. Shelby@shelbynewsad
2 new business models in the discovery time in 5 years: - neolabs for bio - next-gen CROs that are selling proprietary data to frontier labs and biotechs many questions about moats (neolabs) and durability (CROs) but ramp to 9-figure revenue is tangible and relatively rare in biotech.
English
Matthew Osman retweetledi
Really looking forward to moderating this incredible panel! Join us to learn more about the data generation gap in biology and why this is the right moment in time to think about it.
John Cumbers@johncumbers
We can sequence a single cell. We can map every protein in a tumor. We can perturb thousands of genes at once. So why can't we predict what a drug will do in a patient? #SynBioBeta2026 is May 4-7th in San Jose, California, you can learn more about the conference and get your tickets here: syntheticbiologysummit.com/?utm_source=X&… The bottleneck isn't data. It's emergence. Molecular measurements don't automatically tell you what happens at the cell level, the tissue level, or the organism level. Predictive power breaks down at every scale transition, and nobody has fully closed that gap yet. Building the virtual cell, and eventually the virtual organism, requires confronting where our models actually fail and why. At SynBioBeta 2026, a session digs into exactly that: Johnny Yu (CSO & Co-founder, Tahoe Therapeutics), Gleb Kuznetsov (Co-founder & CEO, Manifold Bio), Micha Breakstone (Co-founder & CEO, Cellular Intelligence), and Fabio Boniolo (CSO & Co-founder, Polyphron). Tahoe built the world's largest single-cell perturbation atlas and runs pooled in vivo drug screening across hundreds of patient-derived models in a single experiment. Manifold tests millions of biologic variants directly in living systems and just closed a $55M upfront deal with Roche. Cellular Intelligence generates dynamic perturbation data at 1,000x conventional efficiency to train a universal cell-signaling model. Polyphron is using AI trained on developmental biology to engineer functional human tissue from scratch. Four companies, four angles on the same unsolved problem. The session runs May 6 from 3:30–4:15 PM in the AIxBIO track. If you work in AI-driven drug discovery, virtual cell modeling, or the data infrastructure that makes any of this possible, this is the session to be in.
English
I'll be giving a spotlight talk at SynBioBeta 2026 on May 5th at 2:10 PM on the Main Stage — 'From Cells to Simulation: Building the Data Engine for Predictive Human Biology.' I'll be sharing how iPSC-derived tissues can close the data gap standing between us and a true biological simulator. See you there! #synbiobeta2026
English
Super excited to give this talk at @SynBioBeta
John Cumbers@johncumbers
A general-purpose biological simulator, one that can predict how the human body responds to any intervention, isn't blocked by compute. It's blocked by data. Not data in general. Causal, human-relevant data. The kind where molecular interactions actually produce functional outcomes you can measure. @thematthewosman co-founded Polyphron on the premise that iPSC-derived tissues are the answer: living human tissue that functions as both a therapeutic platform and a scalable engine for generating that ground-truth data. Last week they published results showing their platform converged on high-performing tissue across every genetically diverse donor tested, cutting optimization burden from hundreds of conditions to tens, even with 2,000+ differentially expressed genes between lines. Matt presents at @SynBioBeta 2026, May 4-7 in San Jose. You can learn more about the conference and get your tickets here: syntheticbiologysummit.com/?utm_source=X&…
English
Matthew Osman retweetledi
An outstanding result, by any measure!
I wrote about this compound, daraxonrasib, in 2024.
research.dimensioncap.com/p/rmc-6236-a-c…
We've only scratched the surface of small molecule pharmacology. This drug is a macrocyclic glue, and it achieved something spectacular. Let's keep going. :)
Anirban Maitra@Aiims1742
🚨🚨🚨 RASOLUTE-302 Ph3 is POSITIVE "Daraxonrasib demonstrated a median OS of 13.2 months versus 6.7 months for chemotherapy, with a hazard ratio of 0.40 (p < 0.0001)".... WOW! AMAZING news for patients with #PancreaticCancer The RAS Revolution is ON!! ir.revmed.com/news-releases/…
English
Matthew Osman retweetledi
Our Long-Term Benefit Trust has appointed Vas Narasimhan to Anthropic's Board of Directors.
Vas brings more than two decades of experience in medicine and global health, including as CEO of Novartis.
Read more: anthropic.com/news/narasimha…
English
As biology AI models mature, biology-native data infrastructure will be critical in separating the next generation of winning biotechs. It will come down to three things: biology-native data at scale, agentic AI across R&D workflows, and closed-loop lab automation.
Bessemer@BessemerVP
The next generation of biotech winners won't be defined by their science alone. They'll be defined by their data infrastructure. 🧬 Bessemer's Biotech team is out with their roadmap on biology-native data infrastructure + the market map of the companies building it: → Multi-modal datasets built for drug biology → Agentic AI across entire R&D workflows → Closed-loop lab automation Read the full report: bvp.com/atlas/building…
English
totally agree ;)
Bessemer@BessemerVP
The next generation of biotech winners won't be defined by their science alone. They'll be defined by their data infrastructure. 🧬 Bessemer's Biotech team is out with their roadmap on biology-native data infrastructure + the market map of the companies building it: → Multi-modal datasets built for drug biology → Agentic AI across entire R&D workflows → Closed-loop lab automation Read the full report: bvp.com/atlas/building…
English
The first company to biological super intelligence is the one that can credibly make clinical trials capex (industrial scale data generation with increasing returns) instead of opex (one time validation of a hypothesis with no transfer learning) and realize the financing benefits that would imply.
Matthew Osman@thematthewosman
English
@rlsayar But (1) perfusion tech isn’t robust enough for this and more importantly (2) near unsolvable (practically and ethically) supply constraints prevent acquisition of population scale tissue samples across multiple tissues let alone multiple tissues from same patient
English
@theodorus5 we’ll be publishing more detailed things about the technical roadmap over the coming weeks - stay tuned!
English
@thematthewosman Replicating all the essential features of the developmental endogenous microenvironment(s) to the fidelity required ex vivo is NOT going to be easy tho I must say... would be very interested to hear how you're planning on doing that if that's something you are willing to share
English
Superintelligence won't excuse us from running the experiments.
A frontier AI can read every paper ever written about biology and it still can't predict what a drug combination does to cardiac tissue from a genotype it's never seen. The causal, interventional data frontier AI systems will need to learn from does not exist yet.
We wrote up why the binding constraint on AI solving biology won't be intelligence. It'll be the data. And where that data has to come from.
Matthew Osman@thematthewosman
English
@theodorus5 100% agree we don't think printing is the answer here
English
@thematthewosman Now if you had said you were planning on micromanaging a 4D tissue printing scheme (or worse a 3D one...) I would have been *very* sceptical but 'iteration' 'closed loop' and 'models' etc sound MUCH more like a winning strategy IMHO🙂
English
We have a lab in the loop set up that uses rich multimodal developmental atlases as reference then models design interventions to get iteratively closer to that reference - this closed loop allows us to use active learning and other techniques to get closer and closer to native tissue development
English
@brunogiotti There are a few ways to counteract this, we back test these tissues against clinical trial response to calibrate against that. For example, our cardiac tissues had the same dose response to drugs that are known to cause or not cause arrhythmias
English
@thematthewosman You assume you can learn some information in ex vivo that can predict in vivo phenotype but it might not be the case?
English
physical AI for biology, here in San Francisco
Michelle Lee@michellearning
The next industrial revolution isn’t software. It’s science.
English