Biohub

The difference between these two photos may not look like much. But to scientists, it’s night and day. The laser phase plate makes it possible to see the tiny details inside cells clearly — structures that may hold the key to understanding how diseases start and how to stop them. It required bouncing a laser 10,000x between perfectly polished mirrors — so perfect that there’s only one team in the world that can do it. The idea for it was first conceived over 15 years ago by University of California, Berkeley/Berkeley Lab physicists Holger Müller and Robert Glaeser. Today, it's real. The clearer we can see inside cells, the closer we get to curing disease. Learn more: https://bit.ly/4efNsQb

A new era of science is here — and it starts with the integration of frontier AI and frontier biology. Priscilla Chan, Mark Zuckerberg, and Alex Rives joined Sarah Guo and Elad Gil on No Priors to go deep on what Biohub is building and why it matters: a world model of protein biology and a vision for what health can look like when AI is part of the equation.

We're hosting the New York Symposium on Immune Cell Engineering and Reprogramming this July—bringing together researchers across immunology, synthetic biology, bioengineering, and AI to explore the next phase of programming immune systems. 📅 July 22–23. Our virtual registration and in-person waitlist are still open. Check out more info: https://bit.ly/4ttMyGh

📣 New pre-print: multimodal perturbation atlas. Imaging + scRNA-seq, 57M cell images. 🧬 🔬 Cells are complex dynamical systems — but most ways we measure them destroy them. We built this atlas to ask: how does live-cell imaging compare to scRNA-seq, the field’s gold standard? The answer surprised us. https://lnkd.in/gJ-efhvX We profiled 1,000 gene KOs in A549 cells across three modalities: • Fluorescence imaging (39 live + 13 fixed markers) • Label-free phase imaging of the same live cells • scRNA-seq via CROP-seq 57M cell images. >600k transcriptomes. 100% open. TL;DR: all methods are amazingly powerful, but label-free phase imaging (i.e. cell morphology) matches — and given enough cells, exceeds — the phenotypic resolution of both fluorescence imaging and scRNA-seq. This lays the foundation for live-cell profiling of phenotypic trajectories.  🚀 Fantastic collaborative work from Biohub team. Special shout-out to Chad Liu, Alex Hillsley Madhurya Sekhar, Cassidy Jones, Gabriel (Gav) Sturm & Taihei Fujimori.

One of the biggest bottlenecks in cryo-electron tomography isn't the microscope — it's sample preparation. Rapidly freezing biological samples (vitrification) is inconsistent and hard to optimize, limiting what we can see inside cells. We're funding two-year projects to change that: 🧊 Develop better methods for freezing cells and tissues 🧊 Create new ways to monitor freezing quality in real-time If you're working in cryogenics, heat transfer engineering, materials science, or related fields — even if you've never worked with biological samples — we want to hear from you. The best solutions often come from unexpected places. Better sample preparation means more high-quality images, faster discoveries, and deeper insights into how cells work and what goes wrong in disease. That's how we get closer to curing it. Learn more and apply: bit.ly/4uip6eJ #CryoET #OpenScience #Biohub #CellBiology

Proteins are the machinery of life. Scientists have cataloged billions of protein sequences—but their biology is still mostly unknown. ESM Atlas is a new way in. 6.8 billion proteins. 1.1 billion predicted structures—the largest application of AI to protein biology to date. ESM Atlas makes the uncharacterized parts of protein space searchable for the first time. And it's fully open. Start exploring: https://bit.ly/4e4xOXV

I’m incredibly excited about the release of ESMFold2, ESMC, and the new ESM Atlas. This was a massive team effort, and I’m deeply grateful to have worked with such an amazing group at Biohub. This system is built on a world model of protein biology trained on billions of sequences across the tree of life. I’m hopeful it can serve as a scientific engine for prediction, design, and discovery — helping researchers understand proteins, uncover the molecular basis of disease, and design new therapeutic molecules. ESMC learns representations of proteins from evolutionary sequences. ESMFold2 uses those representations to predict atomic structures of proteins and biomolecular complexes. ESM Atlas makes billions of sequences and predicted structures searchable, opening up a new way to explore both known and unknown biology. One result I’m especially excited about: ESMFold2 can be inverted to design new protein binders, including antibodies — one of the hardest and most therapeutically relevant design tasks. Using a simple gradient-guided search protocol, we designed minibinders and single-chain antibodies (scFvs) against five therapeutically relevant targets in oncology and immunology: EGFR, PDGFRβ, PD-L1, CTLA-4, and CD45. Across every target and modality, the protocol recovered nanomolar or tighter binders. Scaling inference compute translated directly into better experimental outcomes. By generating more candidates and using more critics for ranking, minibinder hit rates improved from 54% to 70%, and scFv hit rates nearly doubled from 12% to 21%. We characterized designs beyond binding. One designed PD-L1 scFv binds with 4.3 nM affinity, shows target-specific cell-surface staining, and relieves PD-1/PD-L1-mediated suppression of T-cell signaling with therapeutically relevant nanomolar potency in a cell-based assay. What excites me most is the shift this represents: as digital representations of biology improve in their fidelity, more of the earliest search for therapeutic protein binders can move from empirical screening into computation-guided design. Just as importantly, these tools are being released openly. ESMFold2, ESMC, ESM Atlas, and the binder design protocol put powerful capabilities for protein prediction and design into the hands of researchers across academia, biotech, and pharma. My hope is that making this system broadly available will help accelerate the field — enabling more people to explore protein biology, test new hypotheses, discover new mechanisms, and design molecular tools that can help prevent, treat, or cure disease. Binder design notebook: https://lnkd.in/gbisi7Q6 Modal recipe for scaling up design: https://lnkd.in/gsz6TVpy Paper: https://lnkd.in/gwf6NsnB

We released a world model of protein biology: a scientific engine for prediction, design, and discovery that consists of ESMFold2, ESMC, and ESM Atlas. Together, they're helping open up a new way for researchers to design proteins and speed up scientific discovery. Our mission is to cure or prevent disease. To do that, we need to accelerate science. https://bit.ly/3RANAlG