Super excited to announce our recent work was accepted to NeurIPS 2025! 🌟 We introduce Prose, a 280M-parameter transferable normalizing flow proposal for efficient sampling of unseen peptide sequences 😮 Many thanks to the fantastic team!

This is huge! A UCLA team managed to build an optical generative model that runs on light instead of GPUs. In their demo, a shallow encoder maps noise into phase patterns, which a free-space optical decoder then transforms into images—digits, fashion, butterflies, faces, even

Solving the many-electron Schrödinger equation with Transformers Every material property, in principle, comes from solving the many-electron Schrödinger equation. But the math is brutal: the Hilbert space grows exponentially, and even the best methods—DFT, coupled-cluster,

🔧 Introducing AbBFN2: our multi-modal antibody foundation model. AbBFN2 jointly models 45 data modes spanning sequences, genetic information and developability attributes to provide a rich framework with which to define conditional generation tasks. Join Research Scientist

🧬Introducing AbBFN2, our latest generative AI model for multi-objective antibody design!✨ Built on our BFN work published in @NatureComms, AbBFN2 masters the dependencies between sequence, genetic attributes, and developability, taking antibody design to the next level! 🧵

Explore the details further: 📄 Preprint: https://t.co/yyC9A0pCvE 🌐 Try the web app and test-drive AbBFN2 yourself: https://t.co/aEbkHEhAMs 💻 Code: https://t.co/sNnhkWdeaw ✒️ Blog:

AbBFN2 highlights how Bayesian Flow Networks enable "condition anywhere, generate anywhere," transforming antibody design workflows from annotation and prediction to complex optimisation tasks.

🌟 De Novo Library Design: Targeting VRC-01 antibodies—rare variants known for their HIV-suppressing capabilities—we successfully generated human-compatible, liability-free libraries despite limited initial examples (only 21 paired sequences in OAS).

🚀 Joint Humanisation & Liability Removal: Unlike traditional sequential optimization, AbBFN2 efficiently tackles both simultaneously. Achieved ~80% success rate (>90% for solvable cases) while minimising framework mutations.

🔄 VH-VL Interface Redesign: Given one antibody chain, AbBFN2 generates stable partner chains, confirmed by Rosetta-calculated binding energies comparable to natural antibodies—despite only ~75% amino acid recovery. This indicates robust, novel design capability.

🛡️ Immunogenicity Prediction: AbBFN2's humaness logits correlate strongly with observed ADA (anti-drug antibody) responses in clinical data—competitive with state-of-the-art predictive methods. High confidence predictions closely align with real-world outcomes.

🏷️ Sequence Annotation: Predicts genetic and developability attributes from sequence alone, matching specialised methods but with much broader coverage.

We extensively validated AbBFN2 across multiple antibody design challenges... ✅ Unconditional Generation: Produces natural antibodies with well-distributed, correlated genetic and developability traits.

Building on our recent ProtBFN paper in @NatureComms, AbBFN2 uses Bayesian Flow Networks (BFNs) to integrate heterogeneous antibody data into a unified generative framework. A single unconditional training objective unlocks flexible, guided conditional generation at inference.

Excited to announce our latest work: AbBFN2, a generative antibody model co-modelling sequences, genetic origins, and developability attributes across 45 diverse data modalities! 🧬🔬 🧵 Thread to follow, including links to the paper 📄 , code 💻, blog ✒️ and web app 🌐!

AbBFN2: A flexible antibody foundation model based on Bayesian Flow Networks 1. AbBFN2 is a generative foundation model for antibodies built on the Bayesian Flow Network (BFN) paradigm, allowing conditional generation across 45 sequence, genetic, and biophysical data modes

AbBFN2: A flexible antibody foundation model based on Bayesian Flow Networks https://t.co/0uRWQ3UwGE #biorxiv_bioinfo

Our work "Protein Sequence Modelling with Bayesian Flow Networks" is now published in @NatureComms! 🎉 🧵 For a breakdown, see my original thread. 📄 Paper

#Oxford-based #Lumai, an #AI accelerator startup using #optics to address global computational challenges, has secured more than €9.2 million to help AI #data centres reduce costs and boost performance 🇬🇧 🚀 https://t.co/Oyt97wGg0N

To learn more, see the original paper from September and our new blog. 📄 Preprint  https://t.co/Yt4iKOuJ9K ⭐ Code  https://t.co/iP2cfQfNkv ✒️ Blog https://t.co/59DaPC9xkU 🤗  https://t.co/Ca4x4E7BoM There is more coming in the next few weeks…watch this space! ⌛ 👀 🔥