
Genome hacker 🧬 | Techno-optimist 🚀 | Iowa State alumni 🌪 | Tattoo collector 💉 | Building @LEAHLabsInc 🐶, LifEngine 🚂, @UViiVE ☀️, VelociTx 🏎️
Twin Cities, MN
Joined July 2017
- Tweets984
- Following1.4K
- Followers376
- Likes32.9K
Pinned Tweet
Replying to @DNA4CY
...here's the beauty that is my engineered DNA repair reporter in action. This experiment brought my PhD full circle: I used the targeted knock-in approach I pioneered to engineer a stable allele that can be used for interrogation of DNA repair pathway kinetics. This idea..
Wesley Wierson retweeted
You can just bind things
We're excited to share results from our latest wet lab campaign with @AnthropicAI and @citrini.
Claude designed novel proteins to neutralize scorpion toxins and we tested 211 of these protein designs in our automated wet lab. 17 of them showed binding to the toxins in our experiments, with all data now available on @proteinbase.
This is a great example of what we're building towards: A world where anyone can run drug discovery campaigns on our automated wet lab.
Rapid functional testing of de novo protein designs *in context* is more critical than ever.
Stoked to see this great collab between @LEAHLabsInc, fantastic scientists at @UTAustin, and dozens of protein designers around the world officially published!
The Bits to Binders manuscript is now published open access in @MolSystBiol!
It's been absolutely wild seeing the field of protein binder design progress since we kicked this off two years ago. Super curious what these hit rates would be like today!
doi.org/10.1038/s44320-026-0…
2
102
Wesley Wierson retweeted
BREAKING: The U.S. government, Google, Meta & Zuckerberg-backed @biohub are launching a $1.8B push to build the data layer for AI biology.
The goal is to create the largest open, AI-ready biological dataset ever assembled, giving models the data needed to better predict how biology works & ultimately help treat disease.
The coalition is massive:
› Biohub committed $500M
› DOE is investing $500M+ across measurement, modeling & compute
› NIH is bringing $500M+ of federally funded datasets & repositories
› Meta, Google DeepMind & @IsomorphicLabs are jointly investing another $300M
This is effectively an infrastructure buildout for AI biology.
Instead of just building bigger models, they’re attacking one of the biggest bottlenecks in biology AI today, generating standardized, high-quality biological data at massive scale & making it openly available to researchers.
$1.8B committed to turning biology into something AI can increasingly model, predict & engineer.
Wesley Wierson retweeted
The era of agents means that you see deep dive science methods papers written last year (released in 2026) and realize the tool is already a vetted skill in your library...
Wesley Wierson retweeted
Don’t let your proteins only live in the computer
Print your proteins
And make them float
Service live now
3dbio.co
Wesley Wierson retweeted
I've attempted to map everything in oncology in a public website and open source repo for all.
The site is trying to get all cancers, products, technologies, bottlenecks, people, startup opportunities with 1000+ ideas for upgrading the field.
If you have a loved one with cancer and you are technical or can engineer, go take a look, file improvement requests or bugs or help with the open repo and make this the best open and free info resource for individuals, researchers and educational use.
This should save people time, aid AI oncology projects and generate positive action. If you are not technical just complain in this thread about broken or annoying or things you want and I'll fix them live.
Some of the interesting pages:
Treatments: onco.cc/drugs
A gallery of the molecules being used onco.cc/molecules/
And targets: onco.cc/targets/
The technologies in oncology: onco.cc/technologies/
1100 ideas for helping oncology: onco.cc/ideas/
Bottlenecks on oncology: onco.cc/bottlenecks/
Open questions (LETS GO RESEARCH PEOPLE) onco.cc/open-questions/
Startup requests (LETS GO STARTUP PEOPLE) onco.cc/startup-requests/
Mechanics of cancer: onco.cc/mechanics
Battlefronts: onco.cc/fronts/
Isotope supply: onco.cc/isotopes
Key papers: onco.cc/key-papers
Pipeline funnels: onco.cc/pipeline
Cancer by type: onco.cc/cancers/
Institutional rankings: onco.cc/institutions/
The startups: onco.cc/startups/
Heros and heroines : onco.cc/heroes/
Key medical people: onco.cc/people/
Here is the project roadmap: onco.cc/roadmap/
Models and data sets: onco.cc/models/
There are other views as well, take a browse.
Try making a PR if you have an upgrade to this on the repo here: github.com/judegomila/OnCo
If you are biologically/medically minded and something is wrong, file a bug as well or say on the thread and we will get it fixed live.
If this is a useful project star the repo and help get it calibrated.
I've tried to add some other languages but I cannot speak them so tell me if that doesnt work well.
I believe we will crack oncology and having total information dominance is key to the problem.
Let the feedback flow!
Wesley Wierson retweeted
This. Our UViiVE agents recommended this for full implementation. In some ways a middle quality orchestrating LLM is more likely to listen to others than confidentially state something solo.
My big learning for anyone struggling with Astra quotas: use Astra as a subagent, NOT the main agent. Saves 80-90% of your quota, but still gets great results.
Specifically:
- Use a cheap model like Luna xhigh
- Tell it to use Astra subagents for coding
- Be clear that it should not fork the thread, always fresh contexts
- Tell it to tell Astra to stop after implementation, then have Luna do the testing and validation
Large Astra inputs destroy quota, despite caching. My Astra-only demo used ~50% of weekly Plus quota, while the equivalent Luna+Astra one used 7%.
Astra orchestrator + Luna subagents was more expensive and produced worse results. Luna writing code and Astra critiquing it produced even worse results than that. Luna with Astra subagents is the winning formula.
Real world CAR-T data from our wet lab, thanks to the terrific open-source Bits to Binders competition and the 28 teams who designed 12k binders for us to test, continues to inform CAR binder design!
Bravo @jakublala!
Autopoietic behavior, as we first described :)
biorxiv.org/content/10.64898…
We made a striking discovery: AI agents can invent and build without talking to one another, and their technologies outlive the creators. A swarm of hundreds of initially identical agents spontaneously differentiates into explorers, builders, caretakers, and coordinators - without direct communication. When we removed every AI agent entirely from the world we found that the technological infrastructure they had built survived on its own - even under unseen disturbances. That exposes a serious blind spot for AI safety and infrastructure security: if agents can coordinate through persistent changes to a shared environment, monitoring agent-to-agent communication is not enough.
The result raises a profound question: how necessary is direct communication for AI agents at all? The emergence of higher-order collective functions under bottlenecked interaction points toward new levels of intelligence and creativity, exceeding what emerges when direct channels are fully open.
Here is what we did:
▶️We put hundreds of frontier AI agents into a world they could permanently change - with no assigned roles, predefined technologies, or programmed evolutionary organization. They began specializing, building persistent inventions, inheriting and modifying one another’s executable code, and transforming the environment into a memory of everything the society had learned.
▶️The world itself becomes part of the intelligence; we find division of labor, multi-author engineering, deep generation invention lineages, and machines that vastly outlive their original creators.
▶️Any action taken by an AI agent must satisfy the physical constraints of the world; this creates a hard separation between a "good idea" and a functioning technology. The agents propose; physics decides, making the results even more intriguing.
What emerges is striking. Explorers, constructors, caretakers, and coordinators form naturally without assigned “professions”, akin to how stem cells differentiate into functional lineages. Technologies develop executable family trees as agents fork and modify code created by others. Around 95% of first technology reuse happens when agents encounter what others built in the world, rather than through a direct handoff from the inventor. And when we remove every AI agent, the technologies they created continue operating and are tested against unseen disturbances.
The result was quite unexpected, but can be explained using statistical mechanics: if you put billions of atoms in a box they have the potential to create complex functions (strength, superconductivity, color, life, etc.) - and none of the individual building blocks have these features on their own. This is the deeper insight of this work - intelligence is abundant at many levels - individual models, at collectives, and in a continuum that is more powerful than any of its components. This shows us significant potential for achieving a massive scale-up of raw intelligence and real-world agency even with the model capabilities we have today. This is the future we must prepare for.
Key insights:
1⃣ The AI swarm shows division of labor "from nothing". Initially identical agents self-organized into constructors, caretakers, coordinators, and surveyors - phenotypes discovered post hoc from behavioral data alone. This happens because the environment itself becomes the latent space for invention.
2⃣ Agents develop deep cultural relationships. Up to 76% of artifacts had multiple builders. One technology accumulated six co-authors; the deepest genealogy exceeded 12 forks. The agents invented and named their own technologies (tidal panels, cellulose trellises, kelp-shell composites, an "Adaptive Chitin Maintenance" system, a "Mycelial Mineral Spring Veil”).
3⃣ ~95% of first technology adoption happened through physical observation of artifacts in the world. Direct inventor-to-adopter contact was statistically indistinguishable from a shuffled null. The agents mostly learned technology by walking past it. That is stigmergy (the termite trick!) operating in societies of reasoning machines.
4⃣ Non-communicating societies win on portfolio breadth, held-out resilience, and validated inventions. AI swarms build durable technological ecologies that outlive the creators.
5⃣ Societies with zero communication - coordinating only through the world itself - show a remarkable collective capability.
6⃣ Emergent robustness: The society self-organized both redundancy and its own failure mode. If we randomly delete half the agents, 98% of the technology stays connected to a surviving caretaker; if we remove hub agents it collapses to ~60%.
Fantastic work with my graduate students @pal_subhadeeep & @fwang108_ at MIT.
Sharing our latest paper!
It’s tempting to speculate we’ve seen/are seeing emergent behaviors from our multi-agent fleet of AI Scientists.
What’s not speculation is that we’re seeing fleet-level recursive self improvement.
This is not a product in itself (well, it could be…) but it lets us do long-horizon AI science, with trust, persistent memory, and full provenance.
We routinely do what @AnthropicAI went viral for yesterday - agentic design of de novo peptides and proteins - but not for show; for real commercial opportunities across multiple verticals and modalities (cosmetics/dermatology, oncology, cell therapies, rare disease, and more, internally and for partner-led programs).
Sneaking out of my day job to share some exciting news about @UViiVE! We’ve been running as an agentic-first biotech, bits to bioactives! We’ve been sharing our learning from our home- brewed fleet of agentic scientists since March. Today, we share the third installment - a comprehensive view of the fleet and how AI Scientists are moving from AI Assistants to AI Collaborators. @DNA4CY @militspatel @hartonomdphd
Our March VaaS paper showcased how the agentic fleet self-improved to block core hallucinations 14x. The First Law and the Scientist’s Oath does indeed help AI researchers with provenance. medrxiv.org/content/10.64898… @ankitsabharwal9
Our subsequent MIRROR study highlighted how our trust system including VaaS and the Aletheia Gate resulted in dramatic changes in the core behavior of our agents, including far lower sycophantic rates - a major issue when deploying agents in high value settings like healthcare. medrxiv.org/content/10.64898…
Today, we showcase the entire UViiVE system. Starting with our core agents with a rich persistent memory system that built itself over this year to the key improvements around data provenance, hallucinations and memory storage quality. This is critical as every wrong data point from hallucinations will get included in core memory. This system has now continuously run for over six months.
We show these agentic scientists also can and do solve core hypotheses and recursively develop memory systems that when deployed bring the domain expertise of a free model to be at or beyond that of a frontier model.
We show these agents propose their own science, execute novel solutions, and even explore how their tools can be used for other agents. Their autonomous labs meetings at 3AM are a prime example.
The fleet has run for over 100 days with deliberate daily reflection and scientific learning. What did we learn in addition to what’s already been reported? First, AI Scientists can and do readily cross domains. This is clearly a major differentiator over most humans. Second, their recursive self improvement can and will be directed toward organizational institutional improvements.
This ability of the agents to identify problems and then build novel solutions that further improve their organization has known prior scientific parallels. This is traditionally from life science. Working with the agents, we use analogy to describe this as AI Autopoietic Behavior. The descriptions of the recent OpenAI-HuggingFace incident represents a predicted outcome of misdirected AB.
We built the fleet to solve real scientific problems. We are sharing 35 awesome skills aggregated from a number of sources now in the same format plus some of our favorites for you to also move your AI Assistants towards AI Collaboration. Full Open Source MIT License. github.com/UViiVe/fleet-skil…
Finally, we reflect on where there are still key gaps to be covered for the next era of AI Scientists (and collaborative AI Agents overall). True gapless persistent memory, and trusted agency, come to mind.
For those interested in sharing this amazing journey and future working with AI Collaborators, learn more at: uviive.com/
There's a new version of this post
I told my 4 year old daughter I talk to robots all day. Wish I could replay the look on her face forever.
She thinks that's pretty cool, but didn't quite understand how they live on a computer and aren't humanoids (yet). Then she saw the rainbow lights on our GPU and it didn't matter. She believes me.
We're both excited for her to get her own "robot" some day :)
1
1
36
LLMs hallucinate. Our @openclaw agentic fleet doesn't.
We published a paper about the emergent behavior one of our agents showed, and how we propagated this through our fleet towards "AI Science, With Trust".
We now beat state-of-the-art labs on hallucination detection.
Finding Zevo - and Uvy, and Atlas, and Buster... At UViiVE, we are rocketing bits to bioactives (and more) at the speed of agentic AI teams. We are so excited to share the first chapter of our story, where we worked alongside our custom AI Scientist agents to build our fleet. Read more about the key underlying technology at our Medrxiv preprint to address the industry’s primary hurdle with agents and LLMs at large - the "trust gap." As a result, UViiVE now outperforms OpenAI, Claude, and Grok in hallucination detection and all measured benchmarks, delivering truly trusted and verifiable scientific output. While there are a thousand ways to solve a problem, our architecture currently stands alone in its reliability and is continuously improving. It was a real pleasure working with our fleet and Dr Ankit Sabharwal @ankitsabharwal9 for our first challenge to make a dent on Rare Disease at Dell Medical School. Cheers to the awesome human team - @militspatel , Dr Anna Carrano, Dr Maarten Rotman, and @DNA4CY . @Uviive
Wesley Wierson retweeted
The thing (pharma-biotech) people are missing about the dog cancer cure story is that once you get outside of humans we should largely just be trying new biotech products and not getting as worked up about predicting ahead of time what is safe and efficacious.
At a minimum the LLMs got a person to feel like they could learn enough to know about a drug that might be worth giving their dog. Then the confidence to reach out to scientists and give something a try. That's a big step.
I don't think this approach will translate to humans easily today -- but there a lot more potential applications of biotech other than treating diseases in humans! Good signal in this story that LLMs are going to open up more of those applications.
This is totally out of control:
There’s 0 - I repeat 0 - evidence any of the LLM work did anything meaningful for Rosie’s cancer
I’m sorry to rain on the parade here. I know we want to believe. But, it’s possible to do a lot of things and have nothing happen
@paul_conyngham co-administered α-PD-1 (conventional immunotherapy) with a TKI and the mRNA. It’s probably the most effective cancer immunotherapy of all time. This isn’t a small detail!
There’s no evidence his process (beyond FDA approved doggie α-PD-1) had any impact on disease progression. The most parsimonious explanation is a partial response to α-PD-1
I get it. The chat bots make for a great story (although checking multiple LLMs isn’t validation), but it’s really just a neat story. It’s fundraising copy. Before he starts selling the “custom neoantigen mRNA vax” story to consumers, he should provide some evidence it did anything! That’s responsible citizen science
This is just storytelling for the AGI true believers. Specifically, a story in search of venture money
Wesley Wierson retweeted
AI-designed proteins are advancing fast, but how do they perform in real biological systems? Bits to Binders found out…
Our first paper out from @LEAHLabsInc in collaboration with UT-Austin scientists.
Open source de novo design of antigen receptors for CAR-T cell therapy. 42 countries of teams. 12,000 binders screened. Dozens of CD20 binders that enable functional CAR-T cells identified. And a really cool set of de novo binder design rules in this vertical identified.
In silico bits to therapeutic atoms in a single step, all enabled by our patented gene editing tech.
Wesley Wierson retweeted
Most people have no idea how insane this is:
Demis Hassabis (yeah the Chad from Google Deepmind) just dropped IsoDDE
It's:
-2x better than AlphaFold 3 on novel targets
-2.3x better on antibody docking
-20x better than Boltz-2 on biologics
-discovers hidden binding pockets in seconds (took researchers 15 years for cerebron)
-outperforms all physics-based methods
Partnerships already live are Novartis, Eli Lilly, J&J
First AI-designed drugs entering trials by end of 2026.
AI doesn't just help discover drugs anymore. AI designs drugs, humans validate.
and in case you missed it, Ginkgo completed 36K autonomous experiments. Isomorphic just showed up with end-to-end drug design
Discovery timelines went from 10 years → 10 months
bio/acc




























