Bio-curious? Join our bio-data hackathon next month. Find a new biological insight in 48 hours using real data. With @AnthropicAI, @awscloud, @biohub, @OwkinScience, @phylo_bio, & more... SIGN UP NOW! ⬇

Biohub Investigator @james_y_zou built a “virtual biotech”—37K AI agents that analyzed 57K clinical trials and found clues to what makes drugs succeed. @nytimes traces its roots to a collaboration sparked by our scientist John Pak: bit.ly/4hLrpnm

Frontier vision-language models can give convincing answers to biomedical questions—yet they often can’t tell what they’re looking at. The MMBU benchmark at #ECCV26 showed this clearly: models frequently get the answer right while failing to identify the modality, body part, or stain. Now, we’re challenging teams to tackle this perception bottleneck in biomedical AI. 3 tracks. $100K+ in compute and prizes. Oct 1–Dec 31. 📅 Good news: registration has been extended through September 30, so there’s still time to join! Thanks to @gxl_ai, @AnthropicAI, @StanfordAILab, @na2uqi, and @biohub for supporting the challenge! #MedicalAI #MultimodalAI #ComputerVision

What if any dataset, any size, any number of dimensions, opened in a browser tab from a link? 🔬🧪💻 Luxar is out today: write it in Python, share it as a link, explore it in any browser. Open source. 🧵 @biohub

Three webinars on the science behind ESMC interpretability, ESMFold2, and binder design are now on YouTube. Hear from the researchers who built the models, and view live demos and tutorials that you can apply to your own research. Watch the playlist: bit.ly/4h1IObl

@biohub and @AnthropicAI collaborated on these super cool kernels! We contributed some ideas on the ESMFold2 side of things that enabled us to extend the context length of the proteins we can predict on a single GPU, and speed up the kernels for ESMFold2

Anthropic just used Claude to accelerate 30+ open-source biology models, including our own ESMFold2. Our researchers contributed to speed up the kernels and dramatically decrease the memory requirements, enabling much larger protein folds.

The ESMC and ESMFold2 models are now on Huggingface! You can now install it directly from PyPi or from huggingface/transformers v5.16.0. As a part of this, we officially are releasing support for our fused Triton kernels and FoldCP in partnership with @nvidia!

AI has become such a powerful tool, enabling entire areas of science and biology. I got to talk about this as part of a new PBS special on how AI and biotech are transforming medicine. Overall, the documentary is quite interesting! It's worth a watch! pbs.org/video/data-cure-stor…

Biohub scientist @drAOPisco worked with a team of genomics and software experts to more than double the number of cells in the Tabula Sapiens dataset, and increase the number of cell types represented, to 700.

Fluorescent proteins are one of biology’s most useful tools, allowing researchers to track proteins and monitor activity inside living cells. A Biohub research team has now used AI to design entirely new fluorescent proteins, called “Rhobins,” that work across a broader range of organisms and conditions. Tools like these give scientists new ways to observe biology in action and helpo move the field forward much faster.

New in @CellCellPress from Biohub Investigator @BoHuangLab: AI-designed proteins built from scratch (dubbed "Rhobins") work in living cells and match or even surpass fluorescent tags refined over decades of natural protein engineering. Read the paper: cell.com/cell/fulltext/S0092…

Organoids can now mimic human organs, but we study them mostly with static, endpoint assays. A new perspective in @CellBiomat argues for instrumented tissues, which are engineered human tissue woven with dense electrical, chemical, mechanical, and optical sensor arrays. This coupled with open software and digital twin standards will help achieve a faster path from disease mechanism to therapy. Read the piece, co-authored by Csaba Forro, Suji Choi, and others from Biohub: doi.org/10.1016/j.celbio.202…

This is the largest dataset contributed to the Billion Cells Project to date. It is an important step toward the scale and coordination needed to build the Virtual Biology Initiative.

.@alexrives, our Head of Science, and the Biohub team have been named to @TIME's TIME100AI list for 2026. Alex is shaping our work at the frontier of AI and biology, including ESM, our world model of protein biology, and the data to build frontier models through our labs and the Virtual Biology Initiative. Thrilled to see our work + the team behind it be recognized. #TIME100AI bit.ly/4xtHoMo

Tumors aren't one disease—they're patchworks of coexisting cancer cell states, each with its own survival strategy. That's why a single drug often can't keep the disease from coming back. Across two new Nature Genetics papers, we used AI (ARACNe + VIPER) to map these states, and found they're nearly identical across patients with the same cancer. From there, we predicted drug combinations that hit multiple states at once. One pairing doubled survival in a hard-to-treat pediatric brain tumor (DMG), validated in mouse models with ~90% predictive accuracy. bit.ly/45TNIkb

.@alexrives on ESMFold, @biohub's open source system for scientific discovery in protein biology: 1. They folded over 1.1 billion proteins and predicted every single structure. 2. The model learned protein biology so well it hit state of the art on every benchmark they threw at it. 3. They never once trained it on antibodies, but antibody design came out the other side anyway. Experiments that used to mean screening millions of antibodies in a lab can now just run on a computer.

New in @cellcellpress: 15 grand challenges for the field of AI-accelerated biology, inspired by a famous list of 23 math problems created by mathematician David Hilbert. Co-authored by Biohub’s @ShanaOKelley and @califano_lab.

Northwestern and @Biohub scientists have identified new potential drug targets for treating #psoriasis, per @NatureComms. spr.ly/6013B1u21f

Predicting protein structure directly from sequence—no MSA required. Join us Aug 25 at 1pm PT for a webinar on ESMFold2, our open-source structure prediction model. We'll cover the science behind the model, then give a hands-on demo of how to fold proteins and biomolecular complexes, via the API, the app + locally. Save your spot ➡️ bit.ly/4xDQfe9