@BenjMurrell

🇿🇦🇸🇪 Researcher at Karolinska. Viruses. Antibodies. Sequencing. Assaying. Modelling. Computing. Work used to be dry, but is now often wet.

 Karolinska, Stockholm
scholar.google.com/citations?user=I80vy5cAAAAJ&sortby=pubdate
Joined December 2021
  • Tweets246
  • Following186
  • Followers4.4K
70Media
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.
2
11
3
40
5,955
Is anyone else hitting these weird new rate limits?
3
276
Me approving LLM expenses for my lab. xitter.cf/TechCrunch/status/2082…
1
3
173
Who would win in a space battle between GPT 5.2 (Pro), Codex 5.3, Gemini 3 Pro (Preview), and Claude Opus 4.6? Apparently Opus 4.6 and it isn't even close.
4
732
New LLM bench just dropped.
3
368
Big if open.
Isomorphic Labs
@IsomorphicLabsFeb 10
Today we share a technical report demonstrating how our drug design engine achieves a step-change in accuracy for predicting biomolecular structures, more than doubling the performance of AlphaFold 3 on key benchmarks and unlocking rational drug design even for examples it has never seen before. Head to the comments to read our blog.
2
4
43
4,898
If anyone is interested in our Branching Flows work, listen in in just under an hour (9am PT / noon ET / 6pm CET) xitter.cf/HannesStaerk/status/20…
Hannes Stark@HannesStaerk15 Dec 2025
In 1 hour, we will enjoy a reading group session with @BenjMurrell about their "Branching Flows: Discrete, Continuous, and Manifold Flow Matching with Splits and Deletions" arxiv.org/abs/2511.09465 ! Join on zoom at 9am PT / noon ET / 6pm CET: portal.valencelabs.com/stark…
3
6
1,669
Find @charliebtan @majdi_has and I at our poster on PROSE for scalable Boltzmann Generators. Work with Saif Syed, @dom_beaini, @mmbronstein and @k_neklyudov
This Post is from an account that no longer exists. Learn more
1
1
21
2,222
🚨 For those training DL models on proteins, it's possible your "structural" train/test split might have leakage cus tools like foldseek/TMalign (and CATH/SCOP databases) do not always account for structural relationship of circularly permuted proteins:
Aiden Kolodziej@aidenosinetrip13 Dec 2025
Excited to share my recent work, CIRPIN 🐍 with @sokrypton and @MazAbulnaga on the discovery of thousands of naturally occurring circularly permuted protein domains
3
21
1
115
13,696
A technical thread on loss scaling in diffusion and flow matching models (related to a new preprint): Since the dawn of time, people have been messing with (or dropping entirely) these pesky time-dependent loss scaling terms, mostly because the models train better without them.
1
4
1
7
756
more replies
We also show this for Edit Flows, which lies just outside of Generator Matching. xitter.cf/RickyTQChen/status/193…
Ricky T. Q. Chen
@RickyTQChen11 Jun 2025
Padding in our non-AR sequence models? Yuck. 🙅 👉 Instead of unmasking, our new work *Edit Flows* perform iterative refinements via position-relative inserts and deletes, operations naturally suited for variable-length sequence generation. Easily better than using mask tokens.
1
2
305
We're still newcomers in this space, and the literature is vast, so if you think we've missed something or if you have any key citations to suggest, please let us know and we'll be happy to update the preprint!
3
198
Preprint is out! xitter.cf/StatMLPapers/status/19… xitter.cf/BenjMurrell/status/198…
Ben Murrell
@BenjMurrell10 Nov 2025
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.
1
4
408
PDF: arxiv.org/pdf/2511.09465
1
283
We think this will be great for proteins, and it finetunes in. With one GPU we got branching and deletion heads into our protein base model in days. Collab on "BranchingBoltzGen" @HannesStaerk ? xitter.cf/HannesStaerk/status/19…
Hannes Stark@HannesStaerk26 Oct 2025
Excited to release BoltzGen which brings SOTA folding performance to binder design! The best part of this project has been collaborating with many leading biologists who tested BoltzGen at an unprecedented scale, showing success on many novel targets and pushing its limits! 🧵..
2
1
1
8
1,861
The manuscript should be up by tomorrow and I'll drop a link. For us (in the "Department of Microbiology, Tumor and Cell Biology") this is epistemic trespass, so we would much appreciate feedback from the pros @lipmanya @peholderrieth @itai_gat @RickyTQChen @AlexanderTong7
8
705
Load more