Lars Dornfeld

Happy Birthday to our great master student @emma-we.bsky.social - hope you enjoy your life-sized fluorescent protein 💡🥳

New preprint! We measured temperature- and pH-induced aggregation for over 18,000 natural and de novo designed protein domains!

Very happy to be this year’s recipient of the 𝗣𝗲𝘁𝗲𝗿 𝗮𝗻𝗱 𝗧𝗿𝗮𝘂𝗱𝗹 𝗘𝗻𝗴𝗲𝗹𝗵𝗼𝗿𝗻 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗔𝘄𝗮𝗿𝗱 together with @jenspuschhof.bsky.social

www.engelhorn-stiftung.de/forschungspr...

We figured out flow matching over states that change dimension. With "Branching Flows", the model decides how big things must be! This works wherever flow matching works, with discrete, continuous, and manifold states. We think this will unlock some genuinely new capabilities.

New blog post: Skills for the Future

A brief reflection on what skills may (hopefully) remain relevant in the new (hopefully bright) future of LLMs et al.

#AcademicWriting #AcademicChatter

bastian.rieck.me/blog/2025/sk...

OpenFold3-preview (OF3p) is out: a sneak peek of our AF3-based structure prediction model. Our aim for OF3 is full AF3-parity for every modality. We now believe we have a clear path towards this goal and are releasing OF3p to enable building in the OF3 ecosystem. More👇

Are you a PhD student interested in ML and biology or health? Come do an internship with me, @avapamini.bsky.social, Alex Lu, @lcrawford.bsky.social, or Kristen Severson at MSRNE!

Applications are due Dec 1: make sure you include a research statement!

jobs.careers.microsoft.com/global/en/jo...

Quote from the linked article: "Generative AI’s utility in optimizing content for social media algorithms is quite distinct from using real-time algorithms to monitor sensor data about the energy grid. Compressing these numbers into a single category of artificial intelligence is an unfortunate erasure of distinctions that could lead to clearer assessments of generative AI’s impact on productivity."

Yes. Conflating "any data-driven prediction or decision software" with "AI" - a term almost everyone associates with large language models - is a huge gift to the hypemen who claim LLMs will do more energy/climate good than harm.

www.techpolicy.press/generative-a...

Congrats @julianvangerwen.bsky.social and rest of the team 🙌

All that my office requires for now is a good tea preparation station

With this, the last bit of my PhD at @embl.org is finally out!
We developed salad (sparse all-atom denoising), a family of blazing fast protein structure diffusion models.
Paper: nature.com/articles/s42256-…
Code: github.com/mjendrusch/salad
Data: zenodo.org/records/14711580
1/🧵

Really excited to receive the ERC Starting Grant! It’s gonna allow us to do a lot of cool protein design and structural biology method development in the next few years!

Just found this paper during our move. Probably the first one I read about protein design in 2022 😄

We also wrote a little Behind the paper blog post with Lennart and Christian. It gives some idea on what our motivation was behind developing BindCraft and why we chose this particular set of target proteins to work on 🙂

communities.springernature.com/posts/behind...

To briefly summarize my argument (cf. Orwell): When you are being stomped on by a boot, it does not matter whether the wearer is a conscious entity and capable of reflection. It just hurts.

New blog post: Consciousness Is Overrated Anyway (for AI)

In which I, somewhat coherently, try to contribute to the whole "Is AI Conscious" debate.

🔗 bastian.rieck.me/blog/2025/co...

#AI #MachineLearning

📢 New preprint: Aromatic Ring Flips Reveal Reshaping of Protein Dynamics in Crystals and Complexes.

We addressed a long-standing question that people debated already in the 1970s - but it still remained open:
Do proteins in crystals move as they do in solution? We used ring flips to find out.
1/n

Just When I Thought I Was Out They pull me back in! Time to teach machine learning again.

Should we still teach machine learning classes? Let me justify my existence.

RosettaFold 3 is here! 🧬🚀

AtomWorks (the foundational data pipeline powering it) is perhaps the really most exciting part of this release!

Congratulations @simonmathis.bsky.social and team!!! ❤️

bioRxiv preprint: www.biorxiv.org/content/10.1...

The @uwproteindesign.bsky.social's experimental pipeline behind models like RFdiffusion and ProteinMPNN:

- A rapid, scalable, pipeline for producing and characterizing proteins
- A demultiplexing protocol for converting oligopools to clonal constructs

Jason Qian @lfmilles.bsky.social Basile Wicky

Scatter plot of LDDT as a function of sequence identity for high coverage homology models. The horizontal red line at 40% sequence identity highlights the presence high quality models in the low sequence identity region.

Filtering out homologous structures from the PDB at 40% sequence identity is not enough to create a robust test set. Significant leakage persists at this level, and comparative modeling can still produce high quality models.

🙌

Scaling down protein language modeling with MSA Pairformer Recent efforts in protein language modeling have focused on scaling single-sequence models and their training data, requiring vast compute resources that limit accessibility. Although models that use ...

Excited to share work with
Zhidian Zhang, @milot.bsky.social, @martinsteinegger.bsky.social, and @sokrypton.org
biorxiv.org/content/10.1...
TLDR: We introduce MSA Pairformer, a 111M parameter protein language model that challenges the scaling paradigm in self-supervised protein language modeling🧵

Rational engineering of allosteric protein switches by in silico prediction of domain insertion sites - Nature Methods ProDomino is a machine leaning-based method, trained on a semisynthetic domain insertion dataset, to guide the engineering of protein domain recombination.

Our work on rationally engineering allosteric protein switches is now out in Nature Methods: www.nature.com/articles/s41... Thanks a lot to @grunewald.bsky.social and @noahholzleitner.bsky.social for the comprehensive news and views: www.nature.com/articles/s41...

Amazing news! And more than deserved 🙌 looking forward to all your new exciting research - Will be an outstanding lab for PhDs / PostDocs!

Today's #RDKit blog post is a heartfelt plea for clearer communication.
greglandrum.github.io/rdkit-blog/p...

Can an AI model predict perfectly and still have a terrible world model?

What would that even mean?

Our new ICML paper (poster tomorrow!) formalizes these questions.

One result tells the story: A transformer trained on 10M solar systems nails planetary orbits. But it botches gravitational laws 🧵

A good example of unfolding „cat“alysis

Was an amazing talk @delafuentelab.bsky.social 🙌

BioEmu is out! Grateful to have had the opportunity to work with such an incredible team on this project 🤗

I was involved in this project as an early stage master’s student in protein engineering and it sparked everything that came after 🙌

I’m incredibly grateful I got to work with the amazing supervisor @juliankompa.bsky.social and the rest of the team on such a cool project. Check it out!