Biohub releases a world model of protein biology

AI just speedran protein discovery — and the comments are screaming why is nobody talking about this

TLDR: Biohub released free AI tools that can predict protein shapes and help design potential disease-fighting molecules much faster than old methods. The community reaction was a mix of awe and disbelief that such a big medical-AI moment got so little attention — plus jokes that broken software still slows scientists down.

Biohub just dropped a giant open-source science toolkit that promises to map huge chunks of the protein universe and even help design brand-new molecules that stick to disease targets in the lab. In plain English: this could make early drug hunting much faster, turning work that can take months or years into days. That’s the kind of claim that should set the internet on fire — and yet the funniest part of this story is that the community reaction was basically, “Wait… why is this thread so quiet?”

That low-comment drama became the real plot twist. One commenter flat-out called it “some of the most exciting and impactful fields of the next years,” sounding genuinely baffled that people weren’t piling in. Another painted a very unglamorous picture of the field until now: brilliant researchers spending their days wrestling with CUDA bugs and package installs instead of curing disease. It’s a classic tech-community mood swing: one minute, awe at a moonshot that could reshape medicine; the next, dark comedy about scientists being defeated by broken software.

There wasn’t much of a fight in the thread, but there was a vibe: quiet astonishment, nerdy optimism, and a hint of “wake up, people, this is huge.” Meanwhile, swyx rolled in with interviews and a YouTube walkthrough like the unofficial aftershow host for a science blockbuster. The meme-worthy takeaway? The future of medicine may arrive before the Python environment installs cleanly.

Key Points

  • Biohub announced an open protein biology AI system consisting of ESMC, ESMFold2, and ESM Atlas.
  • ESMC is a protein language model trained on approximately 2.8 billion sequences from across all of life.
  • In a same-day preprint, researchers used ESMFold2 to design protein binders against five cancer and immunology targets, with computational searches completed in days.
  • The article says the resulting lab-validated binders showed high affinity, specificity, and stability and had minimal similarity to public database sequences.
  • ESM Atlas organizes 6.8 billion protein sequences and 1.1 billion predicted structures to help researchers explore unannotated biology and learned protein relationships.

Hottest takes

"It’s interesting that there are almost no comments on this" — Frannky
"some of the most exciting and impactful fields of the next years" — Frannky
"She spent most of her time fighting cuda bugs and trying installing packages" — Frannky
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.