#RFdiffusion
Architecture breakdown: RFdiffusion (Rosetta Commons) builds protein backbones by denoising. Random residue frames in, a motif or symmetry held fixed, RoseTTAFold fine-tuned as the denoiser, 50 steps by default. Backbone only: ProteinMPNN writes the sequence. toolome.bio #ProteinDesign #ProteinAI
November 21, 2026 at 4:24 PM
AI-designed proteins made to hold a light-driven metal catalyst

Sources:
https://www.nature.com/articles/s41929-026-01645-1

See other stories with deeper dives at https://sharedsapience.com/century-report/?utm_source=bluesky&utm_medium=post&utm_campaign=tcr-newsroom
November 20, 2026 at 6:42 AM
AlphaFold 3 goes further, modeling interactions between proteins, DNA, RNA, and potential drugs. RFdiffusion lets scientists design entirely new proteins, some experimentally validated. None of this is a cancer cure, but pretending AI isn't advancing biomedical science is absurd. 4/4
November 17, 2026 at 4:20 AM
Not LLMs - do not confound LLMs with bespoke tools like AlfaFold, RFdiffusion etc. LLMs will only result in increased patient deaths
November 14, 2026 at 9:01 PM
Can a $1B launch turn de novo AI protein design into FDA-approved drugs? We unpack Xaira Therapeutics, RFdiffusion, and the realities of de novo biology on Target Space.

Listen: bit.ly/484ZbOZ #Biotech #AIDrugDiscovery
November 6, 2026 at 10:12 PM
Can a $1B launch turn de novo AI protein design into FDA-approved drugs? We unpack Xaira Therapeutics, RFdiffusion, and the realities of de novo biology on Target Space.

Listen: bit.ly/484ZbOZ #Biotech #AIDrugDiscovery
November 6, 2026 at 10:12 PM
Guided by these predicted interfaces, we used RFdiffusion to design proteins that specifically bind cofactors. Using a functional screen in cells, we identified more than 300 binders across 10 cofactors! These binders allow us to recruit cofactors individually or in combination.
October 21, 2026 at 7:44 PM
Maybe! David Baker, for ex., is still using his RFdiffusion model in his latest papers. I think an off-the-shelf frontier model is simply just not going to be competitive (at the moment), though it does allow investment firms to play bioengineer. www.nature.com/articles/s41...
Computational design of metalloproteases - Nature
RFdiffusion methods are used to de novo design proteins with water-activating zinc-based active sites adjacent to a peptide substrate binding pocket, yielding highly active, programmable proteases wit...
www.nature.com
October 21, 2026 at 7:19 PM
タンパク質を構成するアミノ酸同士の結合を切断するプロテアーゼを「RFdiffusion」というタンパク質設計用生成モデルで作りました,という研究。
病気の原因になるタンパク質を選択的に分解したり,細胞内のシグナル伝達を操作したりと様々な使い道が生まれそう。

www.nature.com/articles/s41...

de novoなので,元々存在する酵素の改良ではなく,白紙の状態からの設計。こういう時代になったのかと改めて思います。

ちなみにRFdiffusionは2024年にノーベル化学賞を受賞したDavid Bakerのラボで開発されたAIモデル。本論文のラストオーサーも同氏です。
Computational design of metalloproteases - Nature
RFdiffusion methods are used to de novo design proteins with water-activating zinc-based active sites adjacent to a peptide substrate binding pocket, yielding highly active, programmable proteases wit...
www.nature.com
October 20, 2026 at 10:59 AM
1/ The problem is resolution. An enzyme works because a handful of atoms sit in exactly the right places. Earlier tools, including RFdiffusion 1 and 2, mostly sketched the protein backbone first and fitted the chemistry later. Misfit pockets and wrong orientations were common.
October 21, 2026 at 2:05 AM
Protein is the largest category on Toolome: 54 of 293 tools. MIT 20, Apache-2.0 12, BSD-3-Clause 9, other 13; 2024 had the most first releases (11). Featured: ProteinMPNN, ESM-2 / ESMFold, RFdiffusion, BindCraft, Boltz. toolome.bio
October 18, 2026 at 3:23 PM
プロテアーゼのde novoタンパク質設計

AIを用いてアミド結合を切断する高活性な亜鉛メタロプロテアーゼの設計に成功。活性中心に金属イオンや水分子などを精密に配置するのが重要らしい。TDP-43やアミロイドβなどの標的タンパク質を特異的に切断することにも成功
www.nature.com/articles/s41...
Computational design of metalloproteases - Nature
RFdiffusion methods are used to de novo design proteins with water-activating zinc-based active sites adjacent to a peptide substrate binding pocket, yielding highly active, programmable proteases…
www.nature.com
October 19, 2026 at 10:00 AM
We used RFdiffusion, ProteinMPNN, AlphaFold2 filtering, then automated patch clamp as the experimental screen, with no binding assay in between. From ~12,000 backbones and ~70,000 sequences, 76 were expressed and two turned out to be nanomolar inhibitors.
2/n
October 9, 2026 at 8:53 AM
RRIDs were included in this in Nature Communications paper. RRIDs improve reproducibility in scientific research. #RRID #accelerateopenscience #STMpublishing
Improved protein binder design using β-pairing targeted RFdiffusion - Nature Communications
This study demonstrates the capability of deep learning protein design models in generating functionally validated β-strand pairing interfaces, expanding the structural diversity of de novo binding proteins and accessible target surfaces.
doi.org
October 8, 2026 at 7:02 AM
OpenAI’s dot is a protein designer now 😂 It runs RFdiffusion/MPNN/AF2 on its cloud computer, tracks progress in a scratchpad, and opens the predicted designs in PyMOL to click through. All on CPU. Time to enter the Anthropic × Adaptyv competition with this setup? Wish me luck 😆
September 30, 2026 at 7:37 PM
I set up a website to make it easy to run common structural Bio/ML tools like RFDiffusion, structure prediction, and sequence prediction tools (ESMFold, ProteinMPNN, Chai-1 etc). athanortools.com

I'm using this for a specific protein design project and learning journey.
Athanor Bio Tools
Athanor Bio Tools provides small web interfaces and JSON APIs for computational biology software.
athanortools.com
September 19, 2026 at 4:56 PM
Yeah that was pretty good but I was referring to projects like Alphafold and RFdiffusion. Far FAR bigger cans of worms, they're the really scary ones.
September 16, 2026 at 6:47 PM
Researchers developed a two-phase approach combining Rosetta-based motif grafting with deep-learning tools, including RFdiffusion, ProteinMPNN, and AlphaFold2 to generate high-affinity binders that successfully targeted bone regeneration. Read more: rosettacommons.org/2026/09/16/t...
September 16, 2026 at 6:16 PM
TU Munich researchers used RFdiffusion AI to create synthetic transfer vehicles (STV-C8) that outperform LNPs by >1,000-fold in RNA delivery and achieve in vivo CRISPR editing in pigs. 🧬🤖 #Nature2026 #RNADelivery #RFdiffusion #CRISPR #GeneTherapy #STVC8
September 11, 2026 at 9:57 AM
Nice new resource for getting into machine learning for biology 🧬 BioTorch has 116 hands-on PyTorch exercises covering models like AlphaFold2, ESM2, and RFDiffusion. Implement their basic building blocks, test your code, and see how the models work. It is free during beta!
biotorch.chrishayduk.com
BioTorch · Build intuition for biological AI
BioTorch is a practice workspace for the PyTorch building blocks of biological AI. Implement tensor operations, test values and gradients, and return to adaptive reviews.
biotorch.chrishayduk.com
September 7, 2026 at 3:31 PM
New RFDiffusion-for-peptide (RFpeptide) paper from @gauravbhardwaj.bsky.social and team at @uwproteindesign.bsky.social! 🌟 Beautiful binding data on 4 highly-structured targets (pLDDT > 90)! 🙌🏾 Not too confident this would work on highly disordered targets, though. 🤔

www.biorxiv.org/content/10.1...
Accurate de novo design of high-affinity protein binding macrocycles using deep learning
The development of macrocyclic binders to therapeutic proteins has typically relied on large-scale screening methods that are resource-intensive and provide little control over binding mode. Despite c...
www.biorxiv.org
November 20, 2024 at 1:03 PM