AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

La-Proteina: Atomistic Protein Generation via Partially Latent Flow Matching

Tomas Geffner, Kieran Didi, Zhonglin Cao, Danny Reidenbach, Zuobai Zhang, Christian Dallago, Emine Kucukbenli, Karsten Kreis, Arash Vahdat

ICLR 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Generative protein structure design

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add triangular multiplicative update layers (ablation)La-Proteina · Unconditional all-atom generation, lengths 100-500all-atom co-designability (%): 68.4 -> 75.0 co-designability; 206 -> 129 structure clustersMixed
model alpha-carbons explicitly instead of encoding them in the latent space (partially latent vs fully latent) (ablation)La-Proteina · unconditional generationco-designabilityImproved
use separate interpolation times t_x and t_z instead of a single coupled time (ablation)La-Proteina · unconditional generationco-designabilityImproved
generate alpha-carbon backbone at a faster rate than latent variables (ablation)La-Proteina sampling · unconditional generationco-designabilityImproved
reduced noise scaling during sampling (ablation)La-Proteina · unconditional all-atom generation, lengths 100-500co-designability vs diversity: 68.4 -> 60.6 -> 53.8 all-atom co-designability across the three non-tri variantsMixed
unindexed instead of indexed motif scaffolding for motifs with 3+ segments (ablation)La-Proteina · Atomistic motif scaffolding (26 tasks)tasks solvedImproved
use La-Proteina instead of Protpardelle for atomistic motif scaffoldingProtpardelle · Atomistic motif scaffolding (26 tasks)tasks solved: 4 -> 21 to 25 of 26 tasksImproved
add beta-sheet filter to training data (ablation)La-Proteina · Foldseek AFDB training setbeta-sheet content: 3-4% -> approximately 10%Improved
scale La-Proteina to long proteins with 46M-sample AFDB subsetLa-Proteina · Long protein generation up to 800 residuesco-designability and diversityImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy