Pith. sign in

REVIEW 4 major objections 4 minor 3 cited by

ProxelGen: Generating Proteins as 3D Densities

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims protein structures can be generated directly as multi-channel 3D density grids, producing samples more novel and better matched to the training distribution than atomistic point-cloud models at comparable designability.

desk verdict The proxel representation and its spatial conditioning are the real contributions; the unconditional designability claim is currently entangled with a Proteina refinement step not applied to baselines. read the letter →

arxiv 2506.19820 v1 pith:XQGBFMO7 submitted 2025-06-24 q-bio.BM cs.LG

Felix Faltings , Hannes Stark , Regina Barzilay , Tommi Jaakkola This is my paper · ORCID
classification q-bio.BMcs.LG
keywords proteinstructuregeneration3Ddensityrepresentationvoxelgridslatentdiffusionflowmatchingmotifscaffoldingshapeconditioningproxels
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper sets out to show that the representation used for protein structure generation need not be atomistic coordinates: a protein can be generated as a 3D density grid while matching or beating point-cloud-based models. The authors build ProxelGen, a latent diffusion model over a voxelized density representation with Gaussian-smoothed backbone channels and a vector field that encodes chain ordering from the N- to the C-terminus. On unconditional generation, ProxelGen's samples are more novel and closer to the training distribution than a leading atomistic baseline, with roughly native-level designability on the metrics they report. The same representation makes spatial conditioning natural: in motif scaffolding it solves multi-segment tasks where atomistic baselines find one success and ProxelGen finds twelve. If the density representation is what is delivering these results, protein generation gains a new axis—spatial inpainting and shape conditioning without fixing protein length—rather than a marginal tweak to existing point-cloud models.

What carries the argument

The load-bearing object is the 'proxel' representation: a 7-channel 3D voxel grid in which three channels sample Gaussian-smoothed densities around backbone C, Cα, and N atoms, one channel encodes bond midpoints, and three channels form a 'chain flow' vector field pointing along successive Cα–Cα vectors from the N to the C terminus. That chain-flow channel is the mechanism that carries ordering information, making it possible to thread a single amino-acid chain through the generated density. On top of this representation, ProxelGen compresses 32×32×32 proxel arrays with a 3D CNN VAE into a latent space (512× spatial compression), trains a stochastic-interpolant flow model to generate latents, and fine-tunes an atomistic flow decoder to convert generated latents back to backbone coordinates. Because conditioning inputs are also voxel grids, spatial constraints can be injected by channel-wise concatenation in latent space.

What would settle it

Decode the generated proxels with the small coordinate decoder and score designability without the large refinement pass; if designability falls to near zero, the density representation itself is not carrying the reported result. Separately, count the connected components of the generated chain-flow field before decoding: if most samples split or merge, the ordering channel is not doing the threading job the method requires.

Watch

Extended reading notes

Core claim

The paper's central claim is that a protein structure can be generated as a 3D density rather than as an atomistic point cloud, and that this alternative representation is not merely viable but advantageous. ProxelGen encodes proteins as 'proxels'—multi-channel voxel arrays sampled from Gaussian-smoothed densities around the backbone atoms, plus a bond channel and a vector field that traces the chain from N- to C-terminus—and learns a 3D-convolutional VAE whose latent space is generated by a flow model. On unconditional generation, ProxelGen reports higher novelty, better FID against the training distribution, and designability at roughly native levels when compared with a leading atomistic flow model, and in motif scaffolding it reports 12 unique successful designs for a four-segment motif where every tested baseline finds one. The paper also demonstrates that spatial conditioning follows for free from the representation: masked regions and arbitrary voxelized shapes can be concatenated as inputs, enabling inpainting and shape-conditioned generation without prescribing protein length or the placement of motif segments.

Load-bearing premise

The argument assumes that the generated proxels—not the pretrained coordinate decoder and refinement model—are what make the sampled structures designable, and that the chain-flow channel reliably encodes a single connected chain; if either fails, the headline comparisons describe the whole pipeline rather than the density representation.

Editorial extensions

If this is right

  • Protein length no longer has to be fixed before sampling: the same voxel grid can represent different numbers of residues, so generation and inpainting can produce whatever size the conditioning shape supports.
  • Spatial tasks that require awkward bookkeeping in atomistic models—motif scaffolding, masked inpainting, shape-conditioned design—become native operations of concatenating or masking voxel channels.
  • If the 1BCF and 1QJG results generalize, density-based sampling will be the preferred tool for scaffolding multi-segment motifs, where sequence-anchored methods effectively fail.
  • The fixed grid lets protein generation borrow mature 3D CNN and latent diffusion infrastructure from image generation, and the representation itself scales with grid resolution rather than chain length.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A decisive attribution test would be to score designability of decoded structures before the 200M-parameter refinement pass; if designability collapses without refinement, the headline result belongs to the decoder/refinement pipeline rather than to the density representation.
  • If chain-flow connectivity is enforced as a training objective, density-based models should extend to multi-chain complexes and to conditioning on experimental density maps, where a single connected chain is not the right prior; the paper's current chain flow assumes one chain and the authors note it often splits or merges.
  • The shape-conditioning setup suggests a direct application the paper leaves untested: conditioning on low-resolution experimental envelopes rather than shapes derived from known structures, using the same shape-adherence metrics.
  • The fixed-resolution proxel grid also points toward a coarse-to-fine generation scheme—generate a low-resolution shape, then sample higher-resolution refinements—which atomistic representations cannot express naturally.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper introduces ProxelGen, a generative model of protein structure built on a voxelized 3D density representation called proxels. A protein is encoded as a multi-channel voxel grid containing backbone-atom Gaussian densities, a bond channel, and a three-channel chain-flow vector field. ProxelGen consists of a 3D CNN VAE that compresses proxels into a fixed-size latent space, a latent flow model trained to generate these latents, and a fine-tuned Proteina-based coordinate decoder that maps generated proxels back to atomistic backbones. The paper claims that on unconditional generation ProxelGen achieves higher novelty, better FID, and designability comparable to the training set, and that its spatial conditioning enables competitive motif scaffolding and shape-conditional generation. The central claim is that density-based protein generation is a viable alternative to atomistic point-cloud representations.

Significance. If the central claim holds, the paper opens a useful new axis in protein structure generation: representing proteins as voxelized densities rather than atomistic coordinates or frames. This representation naturally supports fixed-size latents, convolutional architectures, inpainting by spatial masking, and shape conditioning, and it connects protein structure generation to the mature literature on voxel-based generative modeling. The paper also provides useful validation of a self-supervised proxel embedding (ProxCLR) for FID-style evaluation, with sanity checks against perturbations and cluster removal. The use of external oracles (TM-align, ProteinMPNN/ESMFold self-consistency, FoldSeek diversity) is a strength, since the headline quantities are not defined purely by the model itself. However, the main empirical claims are currently not anchored tightly enough: the designability comparison is confounded by a large pretrained refinement model, the FID metric is ambiguously defined, and the reported differences lack statistical error bars. These issues are fixable and do not invalidate the core idea, but they must be addressed before the claims can be accepted.

major comments (4)
  1. [Section 3.4, Appendix B.1, Table 1] The central claim that ProxelGen achieves 'the same level of designability as the training set' is evaluated after a renoise-denoise refinement step using an unconditional 200M Proteina model at t=0.8, which is not applied to the Proteina baselines. The paper states that this refinement 'vastly improved' designability while moving structures by about 3.35 Å RMSD. Because the decoder and refiner are large pretrained atomistic priors that are not conditioned on the generated proxels, the reported designability may characterize the Proteina-assisted pipeline rather than the density representation. The Native Proxels row (designability 50.39) being close to ProxelGen's 53.13 strengthens this concern. A control in which random or scrambled proxels are passed through the same decoding and refinement pipeline is needed to attribute the observed designability to the generative model rather than to the decoder/refiner.
  2. [Section 4.2, Table 1] All unconditional metrics are computed on 256 samples with no error bars, no multiple seeds, and no statistical significance testing. Several headline differences are small (FID 6.05 vs 7.25 for Proteina 400M (H); novelty 0.73 vs 0.69; designability 53.13 vs 45.70), and the reader cannot tell whether these gaps are meaningful. Please report confidence intervals, multiple seeds, or a statistical comparison for the key metrics in Table 1.
  3. [Section 3.5, Section 4.2, Appendix B.1] The FID reported in Table 1 is not clearly defined. The text states that 'the FID of the generated proxels themselves, before decoding back to atomic coordinates, is 6.81,' but Table 1 reports FID=6.05 for ProxelGen. Since the refinement step affects both designability and FID, it is unclear whether Table 1's FID is computed on raw generated proxels via ProxCLR, on decoded structures via ProteinFID, or on proxelized versions of decoded and refined structures. The comparison with Proteina baselines is only meaningful if the same embedding and the same pre/post-processing are used for all methods. Please define the FID variant precisely and report pre-refinement and post-refinement values separately.
  4. [Section 6] The paper admits that many generated proxels do not form a single connected chain, that the chain flow may split or merge, and that the coordinate decoder often introduces unnatural kinks when threading a chain through such proxels. This is a load-bearing limitation for the claim that the density representation carries sufficient ordering information for full-chain generation. Please quantify how often chain connectivity fails in the generated proxels, and report how many of the designable samples in Table 1 and the successful scaffolding designs in Table 2 required decoder compensation for disconnected chain flow. Without this quantification, the reported results describe the combined proxel-plus-decoder pipeline rather than the density representation alone.
minor comments (4)
  1. [Appendix A.2] In the sentence describing the contrastive learning details, 'stochasic' should be 'stochastic'.
  2. [Section 3.1, Equation (6)] The chain flow vector field is defined using vectors v_i for i=1,...,N-1, but the summation in Equation (6) runs over i=1,...,N; the indexing for the terminal C-alpha position should be clarified.
  3. [Section 3.5 and elsewhere] The acronym 'FID' is used for both the ProxCLR-based proxel FID and the more general Fréchet Inception Distance naming convention; using a distinct term such as 'ProxFID' would avoid confusion with FID computed on atomistic structures.
  4. [Table 2] The merged length rows in the upper half of Table 2 are difficult to parse, especially because ProxelGen reports a single number per task while prespecified methods report one number per length range; the caption should explain the row structure more explicitly.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: ProxelGen's representation, losses, and evaluations are externally anchored; the disclosed Proteina refinement weakens attribution but does not make any claimed result definitionally equivalent to its inputs.

full rationale

The derivation chain is self-contained rather than circular. The proxel representation is an explicit construction from backbone coordinates (Eqs. 1-6), the VAE and flow objectives are standard reconstruction and velocity-matching losses (Eqs. 7-8), and the headline claims are evaluated against external oracles: ProteinMPNN+ESMFold self-consistency RMSD for designability, FoldSeek clustering for diversity, TM-align for novelty, and external baselines (RFDiffusion, Genie2, FrameFlow, Proteina) for scaffolding. The FID metric uses a self-supervised ProxCLR embedding trained by the authors, but App. C validates it behaviorally against perturbations, FoldSeek clusters, and CATH hierarchy removals, so the metric is not defined into existence by the model being scored. The most serious caveat is App. B.1: the reported designability is obtained after renoising and denoising decoded structures with a fixed pretrained 200M Proteina model at t=0.8, which the paper says 'vastly improved' designability; this is a genuine attribution/control concern for the claim that the proxel representation alone yields designable structures, and a scrambled-proxel control would strengthen the paper. However, that is an experimental confound, not a circular reduction: the refiner is a fixed pretrained prior rather than a parameter fitted to the reported metric, and the equations do not reduce to their inputs. The admitted chain-connectivity failure in Sec. 6 likewise limits the method's practical validity but does not indicate circularity. No load-bearing step is carried by self-citation alone, and the central representation claim has independent empirical content.

Assumptions & free parameters 8 free parameters · 6 assumptions · 3 invented entities

The central claims rest on hand-chosen representation constants (grid spacing 1.5 Å, Gaussian width 1 Å, cutoff 4.5 Å), a chain-flow encoding that the paper admits often splits or merges, PCA axis-alignment for rotation handling, a self-defined FID embedding, and a decoding and refinement pipeline built on pretrained Proteina models. No code or data are released, so the representation and its evaluation stack cannot be inspected independently. These are modeling choices rather than fitted physical laws, but they are load-bearing for every headline number.

free parameters (8)
  • Proxel grid spacing = 1.5 Å
    Hand-chosen resolution in App. A.2; determines how much atomic-position information survives voxelization.
  • Gaussian kernel standard deviation = 1 Å
    Hand-chosen smoothing width for the density channels in App. A.2.
  • Gaussian cutoff radius = 4.5 Å
    Truncation radius for proxelization in App. A.1; controls sparsity and approximation error.
  • VAE KL weight beta = 1e-6
    Chosen small for reconstruction fidelity (Sec. 3.2).
  • Latent downsampling factor = 512x total compression, f not stated
    Spatial compression claimed in Sec. 1; exact f and latent channel count not reported.
  • Flow velocity step-size multiplier = 1.5
    Tuned in Sec. 4.1 to improve FID.
  • Timestep sampling mixture = uniform [0,1] mixed with normal centered at t=0.2, sigma=0.1
    Oversamples early timesteps found important for FID (App. A.2).
  • Structure refinement noise level = t=0.8
    Chosen in App. B.1 because it 'gives the largest gains in designability'.
assumptions (6)
  • domain assumption Gaussian-smoothed atom channels plus chain flow capture enough information to recover designable backbones.
    Sec. 3.1 defines the representation; if side-chain and precise local geometry information is lost, inverse folding and RMSD metrics degrade.
  • domain assumption PCA axis alignment gives a canonical frame so non-equivariant 3D convolutions can model the distribution.
    App. A.2: structures are aligned to principal axes with perturbed-structure alignment; unstable for near-spherical proteins.
  • domain assumption The chain-flow vector field uniquely determines chain order.
    Sec. 3.1, Eq. (6); the paper admits in Sec. 6 that the flow 'may split or merge', so connectivity is not guaranteed.
  • domain assumption ProxCLR SimCLR embeddings yield a meaningful metric for FID.
    Sec. 3.5 and App. C; validation is against the authors' own prior perturbation experiments, and the model is trained on the same proxel distribution.
  • domain assumption ProteinMPNN plus ESMFold self-consistency is a valid designability oracle.
    App. A.2 defines designability; standard in the field but imported without discussion.
  • standard math The stochastic interpolant flow framework is a valid generative formulation.
    Sec. 3.2, Eq. (8), citing Albergo et al. 2023.
invented entities (3)
  • Proxel representation (multi-channel voxel density)
    purpose: Replace atomistic coordinates as the generative target; enables spatial conditioning.
    A modeling construct; its sufficiency is validated only through downstream decoding and FID, with no external falsifiable handle.
  • Chain flow vector field
    purpose: Encode N-to-C ordering of the chain in the density.
    Introduced in Sec. 3.1; the paper reports it frequently splits or merges (Sec. 6), so there is no independent evidence of sufficiency.
  • ProxCLR embedding model
    purpose: Self-supervised representation for FID evaluation of proxels.
    Trained by the authors on proxels; FID behavior is validated only against perturbations and cluster comparisons, with no independent benchmark.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ProxelGen: Generating Proteins as 3D Densities." pith.science (2026). https://pith.science/paper/XQGBFMO7

@misc{pith2026250619820,
  author       = {Pith},
  title        = {Pith review of: ProxelGen: Generating Proteins as 3D Densities},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XQGBFMO7}},
  note         = {Machine review of arXiv:2506.19820}
}
read the original abstract

We develop ProxelGen, a protein structure generative model that operates on 3D densities as opposed to the prevailing 3D point cloud representations. Representing proteins as voxelized densities, or proxels, enables new tasks and conditioning capabilities. We generate proteins encoded as proxels via a 3D CNN-based VAE in conjunction with a diffusion model operating on its latent space. Compared to state-of-the-art models, ProxelGen's samples achieve higher novelty, better FID scores, and the same level of designability as the training set. ProxelGen's advantages are demonstrated in a standard motif scaffolding benchmark, and we show how 3D density-based generation allows for more flexible shape conditioning.

Figures

Figures reproduced from arXiv: 2506.19820 by the authors.

Figure 1
Figure 1. Method Overview ProxelGen trains 3 separate models. First, an autoencoder that spatially compresses our proxel-based protein representation into latents. Second, a latent flow model trained to generate latent protein representations, which are obtained from the encoder that compresses a protein’s proxel representation (encoder weights remain frozen while the flow is trained). Third, a coordinate decoder flow model t… view at source ↗
Figure 2
Figure 2. Proxel Representation A protein structure (left) is turned into a proxel representation with 7 channels: N, CA, C channels, a bond channel, and 3 channels for the chain flow. a simple contiguous sequence. For example, ProxelGen finds 12 unique successful designs on a 4 segment motif, as opposed to a single successful design for all other methods. 2 Background Protein Structure Generation. Deep learning approaches fo… view at source ↗
Figure 3
Figure 3. Unconditional Samples Top row: chain flow of generated proxels. Bottom row: decoded atomic structures. 4.2 Unconditional Generation We evaluate the unconditional proxel samples from ProxelGen, as well as the structures reconstructed from the generated proxels. We compare our model against native structures (Native) from AFDB and structures recovered from native proxels (Native Proxels). As a baseline, we compare aga… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Shape Conditioning Given an input shape (top left) specified as a voxelized density derived from an input structure (top right), ProxelGen can sam￾ple a different structure (bottom right) with a similar shape. This example has a shape F1 score of 0.92, a TM Score to th…
Figure 5
Figure 5. Figure 5: Tradeoffs for Structure Refinement: Refining the decoded atomic structures can lead to different tradeoffs between designability and fid depending on the amount of noise that is added. RMSD is computed between the refined structure and the original structure. 0.4 0.6 0…
Figure 6
Figure 6. Figure 6: Shape conditioning: Samples with the highest TM Scores to the original structures have better shape adherence, but there are also many samples with good F1 but low TM score. Designable samples are also evenly distributed with no evident bias. 16 [PITH_FULL_IMAGE:figur…
Figure 7
Figure 7. Figure 7: Perturbations and FoldSeek Cluster Comparisons In subfigure (a) we show the FID between a reference set of PDB structures and a disjoint set of structures to which various perturbations are applied. In subfigure (b) we show the FID between disjoint samples from pairs o…
Figure 8
Figure 8. Figure 8: FID sensitivity to sample diversity. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

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

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

    cs.LG 2025-07 conditional novelty 7.0 of 10

    La-Proteina generates full-atom protein structures and sequences via flow matching over an explicit alpha-carbon backbone plus fixed-size per-residue latents, achieving state-of-the-art co-designability and scaling to...

  2. 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.

  3. InertialAR: Autoregressive 3D Molecule Generation with Inertial Frames

    cs.LG 2025-10 conditional novelty 6.0 of 10

    An autoregressive transformer with inertial-frame tokenization and geometric rotary positional encoding reports state-of-the-art validity and stability on QM9, GEOM-Drugs, and B3LYP, plus strong functional-group-condi...

Reference graph

Works this paper leans on

34 extracted references · 5 canonical work pages · cited by 3 Pith papers

  1. [1]

    Stochastic interpolants: A unifying framework for flows and diffusions

    Michael S Albergo, Nicholas M Boffi, and Eric Vanden-Eijnden. Stochastic interpolants: A unifying framework for flows and diffusions. arXiv preprint arXiv:2303.08797, 2023

    arXiv 2023

  2. [2]

    Building normalizing flows with stochastic interpolants

    Michael Samuel Albergo and Eric Vanden-Eijnden. Building normalizing flows with stochastic interpolants. In The Eleventh International Conference on Learning Representations, 2022

    work page 2022

  3. [3]

    u rgen J \

    Inigo Barrio-Hernandez, Jingi Yeo, J \"u rgen J \"a nes, Milot Mirdita, Cameron L. M. Gilchrist, Tanita Wein, Mihaly Varadi, Sameer Velankar, Pedro Beltrao, and Martin Steinegger. Clustering predicted structures at the scale of the known protein universe. Nature, 622 0 (7983): 0 637--645, Oct 2023. ISSN 1476-4687. doi:10.1038/s41586-023-06510-w. URL https...

    doi:10.1038/s41586-023-06510-w 2023

  4. [4]

    Se(3)-stochastic flow matching for protein backbone generation, 2024

    Avishek Joey Bose, Tara Akhound-Sadegh, Guillaume Huguet, Kilian Fatras, Jarrid Rector-Brooks, Cheng-Hao Liu, Andrei Cristian Nica, Maksym Korablyov, Michael Bronstein, and Alexander Tong. Se(3)-stochastic flow matching for protein backbone generation, 2024. URL https://arxiv.org/abs/2310.02391

    arXiv 2024

  5. [5]

    A simple framework for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. In International conference on machine learning, pages 1597--1607. PmLR, 2020

    2020

  6. [6]

    Chu, Lucy Cheng, Gina El Nesr, Minkai Xu, and Po-Ssu Huang

    Alexander E. Chu, Lucy Cheng, Gina El Nesr, Minkai Xu, and Po-Ssu Huang. An all-atom protein generative model. bioRxiv, 2023. doi:10.1101/2023.05.24.542194. URL https://www.biorxiv.org/content/early/2023/05/25/2023.05.24.542194

    doi:10.1101/2023.05.24.542194 2023

  7. [7]

    E(3)-equivariant models cannot learn chirality: Field-based molecular generation, 2025

    Alexandru Dumitrescu, Dani Korpela, Markus Heinonen, Yogesh Verma, Valerii Iakovlev, Vikas Garg, and Harri Lähdesmäki. E(3)-equivariant models cannot learn chirality: Field-based molecular generation, 2025. URL https://arxiv.org/abs/2402.15864

    arXiv 2025

  8. [8]

    Protein fid: Improved evaluation of protein structure generative models, 2025

    Felix Faltings, Hannes Stark, Tommi Jaakkola, and Regina Barzilay. Protein fid: Improved evaluation of protein structure generative models, 2025. URL https://arxiv.org/abs/2505.08041

    arXiv 2025

Show all 34 references
  1. [9]

    A latent diffusion model for protein structure generation, 2023

    Cong Fu, Keqiang Yan, Limei Wang, Wing Yee Au, Michael McThrow, Tao Komikado, Koji Maruhashi, Kanji Uchino, Xiaoning Qian, and Shuiwang Ji. A latent diffusion model for protein structure generation, 2023. URL https://arxiv.org/abs/2305.04120

    2023 arXiv

  2. [10]

    Proteina: Scaling flow-based protein structure generative models, 2025

    Tomas Geffner, Kieran Didi, Zuobai Zhang, Danny Reidenbach, Zhonglin Cao, Jason Yim, Mario Geiger, Christian Dallago, Emine Kucukbenli, Arash Vahdat, and Karsten Kreis. Proteina: Scaling flow-based protein structure generative models, 2025. URL https://arxiv.org/abs/2503.00710

    2025 arXiv

  3. [11]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems, 30, 2017

    2017

  4. [12]

    Auto-encoding variational bayes, 2022

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes, 2022. URL https://arxiv.org/abs/1312.6114

    2022 arXiv

  5. [13]

    Generating novel, designable, and diverse protein structures by equivariantly diffusing oriented residue clouds, 2023

    Yeqing Lin and Mohammed AlQuraishi. Generating novel, designable, and diverse protein structures by equivariantly diffusing oriented residue clouds, 2023. URL https://arxiv.org/abs/2301.12485

    2023 arXiv

  6. [14]

    Out of many, one: Designing and scaffolding proteins at the scale of the structural universe with genie 2, 2024

    Yeqing Lin, Minji Lee, Zhao Zhang, and Mohammed AlQuraishi. Out of many, one: Designing and scaffolding proteins at the scale of the structural universe with genie 2, 2024. URL https://arxiv.org/abs/2405.15489

    2024 arXiv

  7. [15]

    Flow matching for generative modeling

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling. arXiv preprint arXiv:2210.02747, 2022

    2022 arXiv

  8. [16]

    Flow straight and fast: Learning to generate and transfer data with rectified flow

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. arXiv preprint arXiv:2209.03003, 2022

    2022 arXiv

  9. [17]

    Lu, Wilson Yan, Sarah A

    Amy X. Lu, Wilson Yan, Sarah A. Robinson, Simon Kelow, Kevin K. Yang, Vladimir Gligorijevic, Kyunghyun Cho, Richard Bonneau, Pieter Abbeel, and Nathan C. Frey. All-atom protein generation with latent diffusion. bioRxiv, 2025 a . doi:10.1101/2024.12.02.626353. URL https://www.b...

    2025 doi

  10. [18]

    Assessing generative model coverage of protein structures with shapes

    Tianyu Lu, Melissa Liu, Yilin Chen, Jinho Kim, and Po-Ssu Huang. Assessing generative model coverage of protein structures with shapes. bioRxiv, pages 2025--01, 2025 b

    2025

  11. [19]

    Bronstein, and Jinbo Xu

    Matt McPartlon, C \'e line Marquet, Tomas Geffner, Daniel Kovtun, Alexander Goncearenco, Zachary Carpenter, Luca Naef, Michael M. Bronstein, and Jinbo Xu. Bridging sequence and structure: Latent diffusion for conditional protein generation, 2024. URL https://openreview.net/for...

    2024

  12. [20]

    Pinheiro, Arian Jamasb, Omar Mahmood, Vishnu Sresht, and Saeed Saremi

    Pedro O. Pinheiro, Arian Jamasb, Omar Mahmood, Vishnu Sresht, and Saeed Saremi. Structure-based drug design by denoising voxel grids, 2024 a . URL https://arxiv.org/abs/2405.03961

    2024 arXiv

  13. [21]

    Pinheiro, Joshua Rackers, Joseph Kleinhenz, Michael Maser, Omar Mahmood, Andrew Martin Watkins, Stephen Ra, Vishnu Sresht, and Saeed Saremi

    Pedro O. Pinheiro, Joshua Rackers, Joseph Kleinhenz, Michael Maser, Omar Mahmood, Andrew Martin Watkins, Stephen Ra, Vishnu Sresht, and Saeed Saremi. 3d molecule generation by denoising voxel grids, 2024 b . URL https://arxiv.org/abs/2306.07473

    2024 arXiv

  14. [22]

    High-resolution image synthesis with latent diffusion models, 2022

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models, 2022. URL https://arxiv.org/abs/2112.10752

    2022 arXiv

  15. [23]

    Score-based generative modeling through stochastic differential equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456, 2020

    2011 arXiv

  16. [24]

    Score-based generative modeling in latent space, 2021

    Arash Vahdat, Karsten Kreis, and Jan Kautz. Score-based generative modeling in latent space, 2021. URL https://arxiv.org/abs/2106.05931

    2021 arXiv

  17. [25]

    Alphafold protein structure database in 2024: providing structure coverage for over 214 million protein sequences

    Mihaly Varadi, Damian Bertoni, Paulyna Magana, Urmila Paramval, Ivanna Pidruchna, Malarvizhi Radhakrishnan, Maxim Tsenkov, Sreenath Nair, Milot Mirdita, Jingi Yeo, Oleg Kovalevskiy, Kathryn Tunyasuvunakool, Agata Laydon, Augustin Žídek, Hamish Tomlinson, Dhavanthi Hariharan, J...

    2024

  18. [26]

    Generating highly designable proteins with geometric algebra flow matching, 2024

    Simon Wagner, Leif Seute, Vsevolod Viliuga, Nicolas Wolf, Frauke Gräter, and Jan Stühmer. Generating highly designable proteins with geometric algebra flow matching, 2024. URL https://arxiv.org/abs/2411.05238

    2024 arXiv

  19. [27]

    Watson, David Juergens, Nathaniel R

    Joseph L. Watson, David Juergens, Nathaniel R. Bennett, Brian L. Trippe, Jason Yim, Helen E. Eisenach, Woody Ahern, Andrew J. Borst, Robert J. Ragotte, Lukas F. Milles, Basile I. M. Wicky, Nikita Hanikel, Samuel J. Pellock, Alexis Courbet, William Sheffler, Jue Wang, Preetham ...

    2023

  20. [28]

    Jason Yim, Andrew Campbell, Andrew Y. K. Foong, Michael Gastegger, José Jiménez-Luna, Sarah Lewis, Victor Garcia Satorras, Bastiaan S. Veeling, Regina Barzilay, Tommi Jaakkola, and Frank Noé. Fast protein backbone generation with se(3) flow matching, 2023 a . URL https://arxiv...

    2023 arXiv

  21. [29]

    Trippe, Valentin De Bortoli, Emile Mathieu, Arnaud Doucet, Regina Barzilay, and Tommi Jaakkola

    Jason Yim, Brian L. Trippe, Valentin De Bortoli, Emile Mathieu, Arnaud Doucet, Regina Barzilay, and Tommi Jaakkola. Se(3) diffusion model with application to protein backbone generation, 2023 b . URL https://arxiv.org/abs/2302.02277

    2023 arXiv

  22. [30]

    Jason Yim, Andrew Campbell, Emile Mathieu, Andrew Y. K. Foong, Michael Gastegger, José Jiménez-Luna, Sarah Lewis, Victor Garcia Satorras, Bastiaan S. Veeling, Frank Noé, Regina Barzilay, and Tommi S. Jaakkola. Improved motif-scaffolding with se(3) flow matching, 2024 a . URL h...

    2024 arXiv

  23. [31]

    Jaakkola

    Jason Yim, Hannes Stärk, Gabriele Corso, Bowen Jing, Regina Barzilay, and Tommi S. Jaakkola. Diffusion models in protein structure and docking. WIREs Computational Molecular Science, 14 0 (2): 0 e1711, 2024 b . doi:https://doi.org/10.1002/wcms.1711. URL https://wires.onlinelib...

    2024 doi

  24. [32]

    Hierarchical protein backbone generation with latent and structure diffusion, 2025

    Jason Yim, Marouane Jaakik, Ge Liu, Jacob Gershon, Karsten Kreis, David Baker, Regina Barzilay, and Tommi Jaakkola. Hierarchical protein backbone generation with latent and structure diffusion, 2025. URL https://arxiv.org/abs/2504.09374

    2025 arXiv

  25. [33]

    Ecloudgen: Leveraging electron clouds as a latent variable to scale up structure-based molecular design

    Odin Zhang, Jieyu Jin, Zhenxing Wu, Jintu Zhang, Po Yuan, Haitao Lin, Haiyang Zhong, Xujun Zhang, Chenqing Hua, Weibo Zhao, Zhengshuo Zhang, Kejun Ying, Yufei Huang, Huifeng Zhao, Yuntao Yu, Yu Kang, Peichen Pan, Jike Wang, Dong Guo, Shuangjia Zheng, Chang-Yu Hsieh, and Tingju...

    2024 doi

  26. [34]

    Tm-align: a protein structure alignment algorithm based on the tm-score

    Yang Zhang and Jeffrey Skolnick. Tm-align: a protein structure alignment algorithm based on the tm-score. Nucleic Acids Research, 33 0 (7): 0 2302--2309, 01 2005. ISSN 0305-1048. doi:10.1093/nar/gki524. URL https://doi.org/10.1093/nar/gki524

    2005 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.