Pith. sign in

REVIEW 10 cited by

Proteina: Scaling Flow-based Protein Structure Generative Models

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2503.00710 v1 pith:WA55QP6A submitted 2025-03-02 cs.LG

Tomas Geffner , Kieran Didi , Zuobai Zhang , Danny Reidenbach , Zhonglin Cao , Jason Yim , Mario Geiger , Christian Dallago
show 3 more authors
Emine Kucukbenli Arash Vahdat Karsten Kreis
This is my paper · ORCID
classification cs.LG
keywords proteinbackboneflow-basedguidancemodelsproteinatrainingbackbones
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Recently, diffusion- and flow-based generative models of protein structures have emerged as a powerful tool for de novo protein design. Here, we develop Proteina, a new large-scale flow-based protein backbone generator that utilizes hierarchical fold class labels for conditioning and relies on a tailored scalable transformer architecture with up to 5x as many parameters as previous models. To meaningfully quantify performance, we introduce a new set of metrics that directly measure the distributional similarity of generated proteins with reference sets, complementing existing metrics. We further explore scaling training data to millions of synthetic protein structures and explore improved training and sampling recipes adapted to protein backbone generation. This includes fine-tuning strategies like LoRA for protein backbones, new guidance methods like classifier-free guidance and autoguidance for protein backbones, and new adjusted training objectives. Proteina achieves state-of-the-art performance on de novo protein backbone design and produces diverse and designable proteins at unprecedented length, up to 800 residues. The hierarchical conditioning offers novel control, enabling high-level secondary-structure guidance as well as low-level fold-specific generation.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 10 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Vilya-1: An all-atom foundation model for macrocycle structure prediction and design

    cs.LG 2026-07 conditional novelty 7.0 of 10

    Vilya-1, a unified all-atom diffusion model, nearly doubles success rates for sampling near-native macrocycle ring conformations versus physics and co-folding baselines and transfers to property prediction and topolog...

  2. Chamaileon: Cross-Context Binder Design with Contextualized Modeling and Mixed Sampling

    cs.LG 2026-07 conditional novelty 6.5 of 10

    I3CD plus MoPS sampling produces single binder sequences that AlphaFold-Multimer scores as compatible with multiple conformational or multi-target contexts on the CROSS benchmark.

  3. Spectral Diffusion for Protein Dynamics

    q-bio.BM 2026-07 conditional novelty 6.5 of 10

    Diffusion over DCT spectral volumes of Cα displacements yields fast, temperature-conditioned protein trajectories with RMSF Pearson r of 0.844 on held-out mdCATH.

  4. Accurate structural modeling of chemically diverse molecular interfaces with Vilya-2

    cs.LG 2026-07 conditional novelty 6.0 of 10

    Vilya-2 predicts bound structures of chemically diverse peptides and small molecules at state-of-the-art accuracy using an all-atom diffusion transformer, recovering 59.1% of peptide interfaces to sub-2 Å backbone RMSD.

  5. Variable-Length Generative Protein Design via Generalized Poisson Flow

    cs.LG 2026-07 conditional novelty 6.0 of 10

    Generalized Poisson Flow learns variable protein length via an inhomogeneous Poisson rate plus within-length flow matching, with KL bounds and gains on structure, sequence, motif, and peptide tasks.

  6. Design-CP: Context Parallelism for Design of Protein Nanoparticles

    cs.LG 2026-07 conditional novelty 6.0 of 10

    Context-parallel inference for RFdiffusion 3 enables end-to-end all-atom design of large symmetric protein nanoparticles on multi-GPU hardware without retraining.

  7. Efficient Molecular Conformer Generation with SO(3)-Averaged Flow Matching and Reflow

    cs.LG 2025-07 conditional novelty 6.0 of 10

    SO(3)-Averaged Flow matching with reflow and distillation enables high-quality one-step molecular conformer generation, reporting new SOTA on GEOM-QM9 and strong one-step results on GEOM-Drugs.

  8. Adaptive Multimodal Protein Plug-and-Play with Diffusion-Based Priors

    cs.LG 2025-07 reject novelty 5.0 of 10

    Adam-PnP guides a pre-trained protein diffusion model with multiple experimental data types, using online noise estimation and precision-based weighting, and reports a backbone RMSD of 0.65 Å on one protein.

  9. MolFORM: Multi-modal Flow Matching for Structure-Based Drug Design

    cs.CE 2025-07 conditional novelty 5.0 of 10

    A flow-matching model with direct preference optimization fine-tuning generates protein-binding molecules faster than diffusion baselines, with improved docking scores on the CrossDocked2020 benchmark.

  10. TABASCO: A Fast, Simplified Model for Molecular Generation with Improved Physical Quality

    cs.LG 2025-07 conditional novelty 5.0 of 10

    TABASCO achieves 0.92 PoseBusters validity on GEOM-Drugs with a 59M-parameter non-equivariant transformer, no bond modeling, and post-hoc RDKit bond recovery, while sampling about 10x faster than SemlaFlow.

Pith tools