You find us here:
bsky.app/profile/graete…
bsky.app/profile/graete…
Latest posts 5
Protein design is one of my new favorites :-) thanks to our best friends, #proteins, and to the fun team behind this👇
Super excited to announce our #neurips2024 paper, "Generating Highly Designable Proteins with Geometric Algebra Flow Matching". Amazing cooperation with Simon Wagner, @LeifSeute, Seva Viliuga @_ncyxx, Nicolas Wolf, and Frauke Graeter @graeter_mbmView quoted post on X
linkedin.com/posts/jan-stuh…
Why connecting helices twice if only one connection carries all the force? #collagen's next surprise. Lead by @benerenne, great fun to collaborate with your Doktorvater @CompBioPhys! @mattertolife @HITStudies @ERC_Research
🥳 Design proteins from scratch, with a good helix/sheet balance, quickly and robustly with fancy geometric algebra 🤓 that catches how proteins look like! Stellar team work by Leif, Simon, Seva @_ncyxx and Nico, wonderful @HITStudies Lab project with @JanStuehmer.
Generating Highly Designable Proteins with Geometric Algebra Flow Matching
1. This study introduces GAFL (Geometric Algebra Flow Matching), a cutting-edge generative model for protein backbone design that leverages projective geometric algebra (PGA) and Clifford Frame Attention (CFA) to enhance designability, diversity, and novelty of generated proteins.
2. GAFL’s innovation lies in Clifford Frame Attention (CFA), which extends the AlphaFold2 invariant point attention (IPA) mechanism, incorporating geometric features such as residue frames in projective geometric algebra to create expressive higher-order messages between residues.
3. Unlike traditional IPA, CFA allows GAFL to capture a more realistic secondary structure distribution, with an emphasis on balancing α-helices and β-sheets, matching the natural diversity found in proteins.
4. GAFL outperforms baselines like RFdiffusion, FrameDiff, and FoldFlow across metrics including designability, diversity, and novelty, particularly excelling in producing small, highly designable backbones with diverse secondary structures.
5. Benchmarking on protein datasets such as PDB and SCOPe shows GAFL’s capacity to generate diverse structures more efficiently, achieving high alignment with natural protein structural distributions.
@JanStuehmer @graeter_mbm
💻Code: github.com/hits-mli/gafl
📜Paper: arxiv.org/abs/2411.05238
#ProteinDesign #GeometricAlgebra #MachineLearning #ComputationalBiology #GenerativeModelView quoted post on X
And more of us! It was great to meet you all 🥰 and many many thanks to bring us all together in pretty Brno @BioExcelCoE. @PilarCossio2 @BonomiMax @CG_Martini @HummerLab
Career first! Girls night out with theoretical chem/biophysics profs in Brno!! 😍😍😍
@graeter_mbm @ppxasjsm @CecClementi @ElisaTelisa
Thank you @BioExcelCoE @amjjbonvin for a wonderful conference! 🤓💕View quoted post on X
