Biology is shifting from discovery to prediction, and patients are helping lead the way.
Priscilla Chan joined @CNBC’s @BeckyQuick on how AI is helping us understand the human body, with the goal of curing or preventing all disease, and why rare disease groups are helping pioneer the future of medicine.
Watch the full interview ⬇️ https://t.co/Z2uFMTf1lQ
📢 Funding alert!
@biohub Rare As One Cycle 4 offers $600K grants over 3 years for patient-led organizations focused on rare respiratory disorders, immune dysregulation & sarcomas.
📅 Apply by Dec 15, 2026.
🔗 https://t.co/kRROCkLmNj
🎙️Tune in to Episode 266 of #TheGeneticsPodcast 🚀
This week, we’re joined by Tania Simoncelli, Vice President of Translational Impact and Engagement at Biohub. Listen in to hear how Rare as One, a network of 94 organizations, has helped advance 123 clinical trials, how AI could improve drug repurposing and patient selection in clinical trials, and the story behind the ACLU lawsuit that took on human gene patents.
Available here👇
🎧 https://t.co/DIfYCqndgK
📽️ https://t.co/Qt7FM40Oam
Ask your biological question in plain language, and your agent picks the right Biohub tool and runs it.
For example: starting from Csm3, search ESM Atlas for uncharacterized relatives and show the most promising candidates.
Introducing the Biohub MCP: run Biohub’s state-of-the-art protein models in your AI agent.
ESMC, ESMFold2, and ESM Atlas now work in any AI agent that supports MCP, including @OpenAI’s ChatGPT and Codex, @AnthropicAI’s Claude, and @cursor_ai.
Get started: https://t.co/ymtbPowt8h
The Biohub MCP works in any AI agent that supports MCP. Add the server URL, install the Codex plugin in one step, or add it as a custom connector in Claude or Claude Code.
Biohub runs the inference, so there’s no code to write, no API keys to manage, and no GPUs to provision.
Folding large biomolecular complexes takes models that can handle longer sequences without running out of compute and memory.
Join Biohub, @nvidia, and @AnthropicAI on Nov 18 at 1pm PT for a webinar on long-context structure prediction with ESMFold2.
Register: https://t.co/IdsTtPi1UM
Zuckerberg’s Biohub Wants to Cure All Disease. Here’s What’s Missing
Can AI help cure all disease?
Alex Rives (@AlexRives), Head of Science at @Biohub, says the biggest holdup is likely not the AI but rather the data: we have barely measured how cells behave. In this conversation, he explains the $1.8 billion effort to generate the data needed to build a "virtual cell."
We talk about how a language model trained on proteins started revealing biology on its own, what a virtual cell actually is, and why so much of this money goes to measurement. We also get into zero-shot protein design, what could shorten the 10-year path to a new drug, what custom medicines for a single patient would require, why Alex says biorisk today comes down to the lab rather than the AI, and the idea of a universal immune system.
Biohub is a nonprofit founded by Mark Zuckerberg (@finkd) and Priscilla Chan, with a mission to cure or prevent all disease. It releases its main protein models openly, including the ESM family that Alex started at @Meta's fundamental AI research lab.
CHAPTERS
0:00 Preview
0:42 Introducing Alex Rives and Biohub
1:48 The $1.8 billion announcement
2:52 Alex on the effort to map the cell
3:37 What is a virtual cell?
4:58 The data gap: why proteins are ahead of cells
7:34 ESM: the first large language model for proteins
9:43 Proteins explained: a 20-letter alphabet
10:30 ESM vs. AlphaFold
13:14 How ESM learns biology by filling in the blanks
14:39 From proteins to cells
15:23 What data do we need to generate?
16:55 Transcriptomes and spatial transcriptomics explained
17:44 Why this needs wet labs
18:56 Will AI replace researchers?
21:01 Why drugs take 10 years, and how to shorten that
23:01 The future of custom medicines
23:40 Zero-shot protein design
25:20 A custom therapy for every patient
26:57 From discovery science to engineering science
29:05 What's missing for custom medicines?
30:22 N-of-one therapies and baby KJ
31:46 Do we need more wet labs?
33:42 Why wasn't anyone generating this data?
38:07 How serious is AI biorisk?
39:51 Gene synthesis screening explained
40:51 A universal immune system
43:37 A call to action
KEY TAKEAWAYS
- Alex calls the core problem a "fundamental data gap." AI for proteins was built on about half a century of data, and nothing comparable exists yet for cells.
- A virtual cell is a model that predicts what a cell will do if you change something. The test is whether its predictions match what happens in a real experiment.
- ESM learned by guessing hidden amino acids. Nobody told it about structure or function, but that information emerged in what it learned.
- Zero-shot design narrows hundreds of thousands of computer-generated candidates to about 100 to test in the lab. Alex describes it as 24 hours of computation and seven days of lab work.
- Custom medicines aren't here yet. Therapies still have to be safe and developable, and the infrastructure and regulatory rules weren't built for a drug only one person will take.
- On biorisk, Alex says "today it's not a risk" and that the key barrier is the experimental laboratory, not AI. He says we need gene synthesis screening, and that defensive applications need to stay ahead.
- Most of this isn't built yet. Alex expects a first dataset in about a year and useful models within five, and those are expectations rather than promises.
Biohub’s new Imaging Residency Program embeds 18 leading scientists from 10 institutions in our research teams to tackle hard problems in biological imaging, from cryo-electron tomography to advanced light microscopy and computational imaging.
Meet the Residents: https://t.co/UEjM3PDwnD
Biohub scientists built a model that recognizes sepsis from blood-cell images without being told what to look for.
Remo-ID, trained only on sample-level labels, got 43 of 47 samples right in an early study.
Led by Paul Lebel and team: https://t.co/nai7V3SnYR
The next frontier for AI reasoning could be understanding how human cells behave.
@biohub's @lorigoler explains why new biological datasets could help AI make scientific discoveries:
"The idea is to sort of create this almost an oracle that's sort of like a generalizable reasoning model for the cell and for biology."
400 million people live with a rare disease. Fewer than 5% have an approved therapy. Applications are open for Cycle 4 of Rare As One, funding patient-led organizations working on rare respiratory disorders, rare immune dysregulation disorders, and sarcomas. Apply by 12/15: https://t.co/sE97GSThOY
After reconstructing a 3D volume of a sample from a series of images, Biohub scientists use an image processing technique called subtomogram averaging to determine structures of macromolecules (such as proteins) in situ. With the xLPP, structures approach atomic resolution.
Learn more about the laser phase plate:
https://t.co/icmV93n6iB
New on the cryoET Data Portal: three cryo-electron tomography datasets benchmarking the dual laser phase plate (xLPP). The microscope captured PP7 virus-like particles in an E. coli cell, and results confirm the unprecedented resolution enabled by the xLPP technology.
The portal puts the data in researchers’ hands to explore, evaluate, and build on. Check it out: https://t.co/fdLERNpd5R
The work was led by Biohub’s Vera Janssen and @loicaroyer, in collaboration with the software engineering team. 🐟
Read the preprint here: https://t.co/YsZsb7U4X9
🧵 The #zebrafish imaging data from @ZFINmod is immensely valuable to researchers, but users are limited to browsing it one gene at a time.
gene2fish, a new Biohub tool, lets you compare dozens of genes side by side, saving hours of time. Try it on Zebrahub: https://t.co/OduCcYyFAY
Single-cell atlases produce long lists of marker genes. gene2fish shows where they’re active in the embryo.
It sits on Zebrahub next to Biohub’s cell-by-cell zebrafish atlas, so gene lists and images live in one place. The code is open source and the data is openly available.
Some astute observers have noted that our @biohub API for ESMFold2 doesn't seem as accurate as some other API services. This is because we predict only in single sequence mode.
Now a max accuracy option exists for in the API to be able to quickly fold your favorite protein!