Meynard-Piganeau Barthélémy

Our new preprint, "SoftAlign: End-to-end protein structure alignment," is now on bioRxiv! biorxiv.org/content/10.110…

Protein function often depends on protein dynamics. To design proteins that function like natural ones, how do we predict their dynamics? @HWaymentSteele and I are thrilled to share the first big, experimental datasets on protein dynamics and our new model: Dyna-1! 🧵

@yorodav @Eurostar I had the same issue. Looks like the free option only appear when the train is officially cancelled.

@Eurostar You guys are jokers, I have train later today, on the main page they say we should rebook but when I want to rebook it , it’s not for free but 120 euros on the app/website…

Due to an object on the tracks near Paris Gare du Nord, we are expecting disruption to our services this morning. Please change your journey for a different date of travel.

@street_spriritt @Eurostar Train finally officially cancelled and now changing is free.

@street_spriritt @Eurostar If you find out how to change the date for free please share it here. I ll do the same.

@street_spriritt @Eurostar Same

@Eurostar Website ask to change ticket, for later (for free) however when following the link I get asked 140€ to change the date of my trip. @Eurostar how to proceed.

Extremely clever new NGS tech from Roche 🧬 If it's hard to discriminate between nucleic acids accurately with a nanopore, why not synthesize a new polymer off a DNA template that is easier to sequence? It's an intuitively simple idea, but took *a ton* of creative nucleic acid chemistry + enzyme engineering to design modified NTPs and create polymers of them. It's been very interesting to see Roche move into the NGS market, and time will tell how this stacks up with the growing set of totally orthogonal approaches people are cooking up for sequencing tech. It feels like we are entering another renaissance for biological measurement infrastructure—which is super important. “Progress depends on the interplay of techniques, discoveries, and ideas, probably in that order." - Sydney Brenner (These animations of sequencing tech will never get old to me. What a time to be a biologist!)

We are looking a PhD Student Researcher at Google DeepMind for 2025 summer! Strong publication record and technical skills on machine learning and genomics preferred. Need to be 80% for a few months. Team in London. Looking forward to hear from you!

Reading "Distilling Knowledge in a Neural Network" left me fascinated and wondering: "If I want a small, capable model, should I distill from a more powerful model, or train from scratch?" Our distillation scaling law shows, well, it's complicated... 🧵 arxiv.org/abs/2502.08606

My talk will be titled: "Toward Next Generation AI Systems Beyond Lingual Intelligence". You will hear about #MBZUAI, the PAN (Physical, Agentic, Networked) world model, and the AIDO (AI-driven Digital Organism) FM4Bio models. See you tomorrow! @mbzuai , @genbioai , @llm360

New paper from the lab🎈. Introducing Micro Immune Response On-chip (MIRO), a device that replicates tumors and their microenvironment to better understand responses to immunotherapies (cyan=immune cells, red=cancer, green=CAFs). @IBECBarcelona

An inspirational talk by Michael Jordan: a refreshing, deep, and forward-looking vision for AI beyond LLMs. youtube.com/live/W0QLq4qEm…

A major challenge for CAR therapies in GBM? The suppressive tumor microenvironment (TME). CAR-T, CAR-NK, and CAR-macrophages each struggle against it, but in different ways. Awesome work from @to123w shows why multifaceted engineering is essential for durable efficacy. 🧵👇

1/ 🚀 Today, we are thrilled to announce the launch of GenBio AI, a new startup aiming to develop the world’s first AI-Driven Digital Organism (AIDO)! A new era in biology and artificial intelligence has begun. 🧬✨ Here’s how we’re transforming research with multiscale foundation models and what makes us stand out. 🧵👇

1/ In two back-to-back papers, we present our de novo TRACeR platform for targeting MHC-I and MHC-II antigens TRACeR for MHC-I: go.nature.com/4gcLzn5 TRACeR for MHC-II: go.nature.com/4gj5OQk @HaotianDu4 @ljjchem @SgourakisLab @StanfordBiosci @UPennAllergyImm @CHOP_Research

Meet Willow: Our state-of-the-art quantum chip. It's the first quantum chip to show exponential error reduction as qubits scale, paving the way for large-scale, fault-tolerant quantum computers. Dive in → go.nature.com/3OKVLY6

Balancing Locality and Reconstruction in Protein Structure Tokenizer • The AIDO.StructureTokenizer (AIDO.St) introduces a VQ-VAE-based approach for protein structure tokenization, transforming 3D structural data into discrete tokens. This tokenizer aligns sequence and structure seamlessly, enabling advanced multimodal applications. • Innovation highlight: The model strikes a critical balance between locality and globality in tokenization, enabling superior performance in both structure reconstruction and homology detection tasks compared to existing tokenizers like ProToken and Foldseek. • AIDO.St features a 300M parameter model with an equivariant encoder and an invariant decoder, trained on over 477,000 protein structures. This architecture ensures rotational and translational symmetry in encoded features. • Compared to competitors, AIDO.St achieves a 2% sacrifice in reconstruction accuracy while significantly boosting retrieval and prediction performance, outperforming ESM3 and ProToken in multiple benchmarks. • Structural reconstruction metrics, including TM-score and RMSD, show AIDO.St’s superiority, especially for proteins with larger radii. The tokenizer reconstructs structures with a TM-score of 0.986 for proteins with radii <100 Å. • AIDO.St demonstrates strong alignment with protein language models (pLMs), boosting structure prediction accuracy. It enables a sequence-to-structure pipeline with unprecedented TM-score improvements in CASP14 and CAMEO datasets. • Practical implications: This tokenizer not only advances structure prediction accuracy but also bridges sequence and structure modalities, paving the way for breakthroughs in protein modeling and drug discovery. @ericxing @dasongle @yingtaoluo @barthelemymp 💻Code: github.com/genbio-ai/AIDO 📜Paper: biorxiv.org/content/10.110… #ProteinTokenizer #3DBiology #AIInBiology #GenerativeModels

BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Chemistry with one half to David Baker “for computational protein design” and the other half jointly to Demis Hassabis and John M. Jumper “for protein structure prediction.”

Don't forget to submit an abstract to LEGO next meeting in Toulouse!