Thanks to @NikoMcCarty and the Asimov team for helping me get this out! Hopefully an interesting view into where we are with AI antibody design today. Featuring @boltz_bio BindCraft @ai4s_protenix @AriaxBio @modal @tamarindbio @diffuse_bio @try_litefold @chaidiscovery @nablabio

I think others have done this too but here is a quick look at ipSAE vs ipDAE vs ipTM across folding models ipDAE (BindCraft2) is consistently good, but OpenDDE+ipTM is surprisingly good and this is robust across different thresholds Consensus doesn't help as much as I expected

I was excited to see the new Gemini looking so good! So we tried to sign our company up. One miserable hour later of messing with inscrutable “enterprise” stuff on workspace and gcloud, we just gave up.

Excited to share our third volume of the @_DimensionCap Research magazine! These articles were a blast to write with my research colleagues @SimonDBarnett and @ChemVagabond, and I feel so lucky to get to brainstorm, debate, learn, and write about these topics alongside them both on a daily basis. You’ll probably notice a theme to a lot of these articles that we have written throughout 2026: the chronicles of us diving head first into AI model training, the data needed to feed them, and emergent applications of training at the frontier. One of my favorite articles I wrote this year was centered around the bio data market and why biological training data has such unique considerations compared to text data. Since I published this one back in June, the demand for high-quality, therapeutic-relevant bio data has become even more obvious, and I don’t see this demand slowing down. The impact that AI is and is going to have in bio is becoming increasingly apparent every day, and we can’t get there without the right datasets. Our goal here is to make these topics accessible and interesting to all, no matter if you’re a frontier ML researcher doing this work daily or brand new to the space and just trying to get a sense of it all. We very much hope you enjoy!

Another good and open (Apache 2.0) folding model, with a focus on antibodies. Interesting that they don't compare to Protenix v2. Something is happening behind the scenes here. "Mol-1 is derived from Protenix, developed by the ByteDance AML AI4Science team."

Sweet project.

@ScottGottliebMD has the story on $MRNA. Despite the negligible efficacy, INTerpath-001 should not be viewed as a failure. Vaccines lack the potency for cancer but tailoring treatment to patients makes therapeutic sense (ask a surgeon!). The FDA door is open; what comes next?

Some great progress from @debesh and the @nanograb team! Delivering mRNA to specific cells would be a huge unlock.

There was so much cool bio AI stuff today, the Vitrify blogpost shouldn't get lost. Really amazing performance in this graph!!

Introducing RFD4-Proteina! A monumental advance in protein design! Diverse frontier capabilities across design tasks and biological modalities from the combined strength of the @UWproteindesign RFdiffusion and @NVIDIAHealth Proteina lineage of models. Available open source, through our partners, and as an NVIDIA NIM.

Dynamics: folding & binding of an intrinsically disordered peptide to its protein receptor via an intermediate state. A 5 milliseconds simulation would take >10 GPU years with MD simulation. At 5 microsecond steps, BioEmu2 needs 40 minutes on the same GPU – a >100,000 x speedup!

I didn't know the *other* drives-upside-down fan car, the McMurtry Spéirling (thunderstorm in Irish) is from a company founded in 2016 by a 74 year old Irish billionaire, David McMurtry (RIP), who made his money from metrology! en.wikipedia.org/wiki/McMurt…

chatgpt just out there scratching itches

Is it possible to use folding @ home to improve e.g., ESMFold2 (MIT license!) Basically, just have distributed folding and self-distillation for many many FLOPs? It seems like a very practical use for all those GPUs. foldingathome.org/science/fo…

Interesting idea! Cost is $5k + $270/yr

why is google like this? so confusing (i clicked lamp shades as one of their default gemini 4 examples!?)

We are giving away 1 MILLION LigandForge designs as part of the Adaptyv Proteinbase + Anthropic protein design competition. challenge.ligandai.com