New preprint🚨
Imagine (re)designing a protein via inverse folding. AF2 predicts the designed sequence to a structure with pLDDT 94 & you get 1.8 Å RMSD to the input. Perfect design?
What if I told u that the structure has 4 solvent-exposed Trp and 3 Pro where a Gly should be?
Why to be wary🧵👇
I thoroughly recommend reading all of Cory Doctorow's recent speech on AI skepticism, it's crammed with new arguments and interesting new ways of thinking about these problems https://pluralistic.net/2025/12/05/pop-that-bubble/#u-washington
Introducing gRNAde: our own little "AlphaGo Moment" for RNA design! 🧬🚀
📝: tinyurl.com/gRNAde-paper
Unlike proteins, RNA design has long relied on "wisdom of the crowd" (human experts) or the slow crawl of directed evolution — gRNAde changes that! 🧵👇
Guiding Generative Models for Protein Design: Prompting, Steering and Aligning
Figure 1
Figure 2
Table 1
Guiding Generative Models for Protein Design: Prompting, Steering and Aligning [new]
Reviews methods to guide generative models to design proteins with specific properties, even if rare in training data. Focuses on parameter and fixed-model methods.
Global Analysis of Aggregation Determinants in Small Protein Domains https://www.biorxiv.org/content/10.1101/2025.11.11.687847v1
Re recent AFDB update, in case you wondered:
- most of AFDB is still same original predictions
-new/changed entries were modeled with AF2
- the MSAs are the originals, so should not contain sequences from last few years
De novo design of All-atom biomolecular interactions with RFdiffusion3 www.biorxiv.org/content/10.1...
Protein functional site annotation using local structure embeddings | PNAS www.pnas.org/doi/10.1073/...
RFdiffusion2 is now live!
github.com/RosettaCommo...
You can now design proteins, and in particular enzymes from just partially defined amino acid side chains, and without defining their sequence position or order!
Scaling down protein language modeling with MSA Pairformer
Figure 1
Figure 2
Figure 3
Scaling down protein language modeling with MSA Pairformer [new]
...Pairformer: memory-efficient MSA, bi-directional updates, better evol. signals, outperforms larger models.
There is @aixbiobot.bsky.social who does a similar thing.
PS, I found Vidu to work better for interprolation between images. Example attached:
Structural motif search across the protein-universe with Folddisco https://www.biorxiv.org/content/10.1101/2025.07.06.663357v1
Hello all Protein Cosmos 🧶🧬 followers. A new Protein Cosmos feed had been set up with a new host. You will need to search Protein Cosmos and add the new feed to your account. Apologies and thanks to @blueskyfeeds.com for all their support to get us started. Best luck for the future!
New preprint 🚨--protein language models + MD training ➡️ allosteric networks!
@sonyahanson.bsky.social and I are developing RocketSHP 🚀 for rapid genome-scale inference of local+correlated fluctuations + structure token distributions!
📄: www.biorxiv.org/content/10.1...
💻: github.com/flatironinst...
For a similar pipeline, there is also juv github.com/manzt/juv that extends uv to jupyter notebooks including inline script metadata (PEP 723).
A case study on the challenges of evaluating AI predictions in biology and the implications for published results.
1/2
rachel.fast.ai/posts/2025-0...
The war on science in the US is already having an effect on private sector research like AlphaFold. Bears repeating but the private sector builds on top of things created by academic research for the public good. This hurts everyone.
To the world:
We are fighting back. Our movement has been silenced by the media here—but we are not backing down. This is what our streets looked like across multiple cities. Tomorrow, there will be more of us! Raise a glass to freedom.
—With love,
Your American allies.
Sharing slides for All-atom Diffusion Transformers
- briefly summarises the big ideas and key takeaways
Link - www.chaitjo.com/publication/...
From a colleague in my PhD lab! Chase presents her method OMEGA, a simple, scalable method to assemble 100s-1000s of custom genes from oligo pools using standard lab tools!
#synbio #proteinengineering #OMEGA
🧬 Meet Lyra, a new paradigm for accessible, powerful modeling of biological sequences. Lyra is a lightweight SSM achieving SOTA performance across DNA, RNA, and protein tasks—yet up to 120,000x smaller than foundation models (ESM, Evo). Bonus: you can train it on your Mac.
arxiv.org/abs/2503.16351
I’ve just updated the BioEmu notebook to include the powerful LogMD. Now, you can generate equilibrium ensembles and explore the full ensemble directly in the notebook.
A huge thanks to Alexander Mathiasen for the support! 🙌
🔗 Try it on Google Colab: lnkd.in/gcuqd-fT
Nitrogenases are vulnerable to oxidative damage.
BIOSS researchers have found "Shethna protein II" that binds to the ezyme to protect it, meaning it can function inside O2-producing cells.
Plants could finally be engineered to make their own fertilizer:
www.nature.com/articles/s41...
Le Monde reporting that a French scientist traveling to Houston to attend a conference was denied entry to US after a search of his phone & computer revealed messages critical of Trump's science cuts, "which [says CPB] conveyed hatred of Trump & could be qualified as terrorism". Computer confiscated
The phrase "pathologically disorganized" is used to describe a neural network trained to predict protein structure from sequence embeddings
A post by @ncfrey.bsky.social and @amyxlu.bsky.social on repurposing ESMFold for protein design, featuring one of my favorite phrases in the field ncfrey.substack.com/p/hit-the-vi...
Introducing BixBench, a benchmark for AI agents in bioinformatics, built with ScienceMachine. We've created 53 scenarios with 296 questions testing AI on computational biology challenges. BixBench includes evaluation metrics and an open-source LLM environment.
We trained a genomic language model on all observed evolution, which we are calling Evo 2.
The model achieves an unprecedented breadth in capabilities, enabling prediction and design tasks from molecular to genome scale and across all three domains of life.
We are happy to share that the code and weights of PLACER (formerly known as ChemNet) are now public!
This method allows for rapid evaluation of protein side chain and ligand conformational ensembles, with uses in ligand docking and enzyme evaluation.
github.com/baker-labora...
