Pith. sign in

REVIEW 4 major objections 7 minor 26 references

Challenges and Guidelines in Deep Generative Protein Design: Four Case Studies

T0 review · 4 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Family-tuned score matching and flow matching can generate monomeric proteins that are clash-free, conserve functional residues, stay stable in molecular dynamics, and bind their family-specific ligands in silico.

desk verdict An honest, reusable evaluation protocol for generative protein design, but the abstract's functional-plausibility claims outrun what the self-referential validation can support. read the letter →

arxiv 2411.18568 v2 pith:K7JXHWZY submitted 2024-11-27 q-bio.BM

Tianyuan Zheng , Alessandro Rondina , Gos Micklem , Pietro Liò This is my paper · ORCID
classification q-bio.BM MSC 92D2092C4068T07
keywords deepgenerativeproteindesignscorematchingflowSE(3)equivarianceconservedresidueanalysismoleculardynamicsvalidationprotein-liganddockingstructuralphylogenetics
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 works to show that two types of deep generative models—score matching and flow matching, operating on the three-dimensional rotation-and-translation geometry of protein backbones—can be fine-tuned on a single protein family and then propose new monomeric designs that survive a battery of in silico functional checks. Across four structurally and functionally diverse families (β-lactamases, cytochrome c, green fluorescent protein, and Ras), the generated backbones are clash-free and adopt realistic dihedral angles, the designed sequences keep functionally critical residues while varying elsewhere, and the designs cluster with their own family in structural phylogenetic trees. Molecular dynamics simulations show the designs remain folded and stable under physiological conditions, and blind docking places family-specific ligands in wild-type-like binding pockets with favorable energies. If these results hold, family-targeted generative design becomes a practical early-stage screen, with the paper offering a ten-point protocol for using such models responsibly.

What carries the argument

The load-bearing machinery is the SE(3)-equivariant backbone generator: a neural network that treats each residue's backbone atoms (N, Cα, C, O) as a rigid frame in three-dimensional space and learns to rotate and translate those frames as a whole, trained either by denoising score matching on a diffusion process over rotations and translations or by flow matching along geodesic interpolants between frames. Both training schemes share a common architecture built on a geometry-aware attention layer that is invariant under global rotations and translations, and both are fine-tuned per family starting from pretrained weights. Around this generator sits a multi-stage validation pipeline: a learned sequence-design network proposes candidate amino acid sequences for each generated backbone; a structure-prediction model checks which sequence folds back to the original backbone; homology modeling adds side chains; molecular dynamics simulations at physiological conditions probe stability; and blind docking places family-specific ligands on the full protein surface to test pocket compatibility.

What would settle it

Express and purify a set of top-scoring designs from one family and assay them for folding and for binding or activity against the family ligand; if the designs fail to fold or show no binding, the functional-plausibility claim collapses. A purely computational falsifier would be to withhold an entire subfamily from training and check whether generated backbones still recover that subfamily's conserved residues and pocket geometries—if they only match training members, the family-specific features are memorized.

Watch

Extended reading notes

Core claim

The paper's central claim is that after family-specific fine-tuning, SE(3)-based score matching and flow matching generators produce monomeric protein backbones that are not merely novel but functionally legible: they occupy allowed Ramachandran regions without steric clashes, recapitulate family-specific structural signatures such as the GFP β-barrel and the Ras switch regions, and encode sequences that conserve the residues known to be essential in each family while allowing variability elsewhere. The evidence trail runs through four validations: structural phylogenetics using Qscore and the 3Di interaction alphabet, which places generated structures within their own family cluster rather than intermixed with other families; ten-nanosecond molecular dynamics simulations under physiological conditions, in which generated backbones stay near their starting geometry with wild-type-like radius of gyration and secondary structure; and blind docking, in which family ligands bind at wild-type-like pockets with binding free energies below −6 kcal/mol. The paper also reports that flow-matching samples are more flexible and more diverse, while score-matching samples are more rigid and closer to the training distribution, a trade-off it treats as a design choice. It frames the whole pipeline as a set of concrete guidelines for early-stage de novo design rather than as a final replacement for experimental validation.

Load-bearing premise

The load-bearing premise is that agreement with the experimentally derived family structures used for fine-tuning counts as evidence of functional and evolutionary relevance, so if the generators are memorizing their training set rather than learning chemically meaningful design rules, the claims of functional plausibility would be undermined.

Editorial extensions

If this is right

  • Family-targeted generation, rather than generic de novo sampling, is enough to reproduce family-specific structural signatures and conserved functional residues in the same pipeline.
  • Structural phylogenetic trees built from Qscore and the 3Di alphabet separate generated designs by family more cleanly than sequence-based trees, so structural comparison can serve as a design-validation layer when sequence identity is low.
  • The score-matching versus flow-matching trade-off gives practitioners a dial: score-matching for rigid, conserved scaffolds and flow-matching for diverse, flexible variants.
  • An in silico screen combining geometric plausibility, conservation, dynamics, and docking can be run before any wet-lab synthesis, with experimental validation reserved for the final candidates.
  • Some generated designs reproduce wild-type allosteric behavior, such as KRas switch opening in the GDP state and closing in the GTP state, suggesting the generative models implicitly capture conformational state information.

Reading between the lines

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

  • Because every validation metric compares generated samples against the same family structures used for fine-tuning, the protocol measures family plausibility, not inventiveness; a held-out subfamily test would separate memorization from generalization.
  • The paper's own note that some diversity regions are under-covered implies a concrete diagnostic: train on one Ambler class of β-lactamases and check whether designs recover structural signatures of the other classes never shown during fine-tuning.
  • Applying the same pipeline to cofactor- or assembly-dependent targets would require conditioning the generator on the cofactor or interface; the paper's limitations section indicates that isolated monomer design is unreliable in those regimes.
  • The score-matching versus flow-matching rigidity trade-off suggests a testable stratification: score-matching-derived designs should perform better in binding screens when a rigid scaffold is needed, while flow-matching-derived designs should win when conformational switching is the function.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. This manuscript presents a computational pipeline for early “de novo” protein design based on SE(3) score matching (SM) and flow matching (FM), fine-tuned separately on four protein families: β-lactamases, cytochrome c, GFP, and Ras. For each generated backbone, the authors use ProteinMPNN to propose ten sequences, select the sequence whose ESMFold model best matches the generated backbone by TM-score, and add side-chains by homology modeling with MODELLER. The designs are then evaluated with Ramachandran plots, ConSurf/Rate4Site conservation analysis, structural phylogenetic trees built from Qscore and 3Di distances, 10 ns molecular dynamics simulations, and AutoDock Vina blind docking. The paper reports that the generated structures are realistic and clash-free, cluster by family, preserve conserved functional residues, remain dynamically stable, and form wild-type-like binding pockets; it distills the approach into ten practical guidelines. The main technical contribution is the adaptable multi-metric evaluation protocol and the qualitative comparison between SM and FM behavior.

Significance. If the claims are correct, the paper would provide a useful, reproducible in silico screening protocol and a practical comparison of SM and FM for family-conditioned backbone generation; the public release of code, scripts, and generated samples is a genuine strength, as is the authors' explicit discussion of limitations in Section 4 and Section J. However, the headline claims of functional and evolutionary plausibility rest on comparisons to the same family structures used for fine-tuning, and the “de novo” framing overstates a pipeline that relies on ProteinMPNN for sequence design and template-based homology modeling for side-chains. I agree with the stress-test assessment that the missing memorization control is the decisive weakness: the evaluation cannot currently distinguish learning family-relevant physics from reproducing the training manifold. The paper is therefore a useful case study and guidelines contribution, but its current evidence does not support the abstract's functional-relevance claims without substantial additional controls or substantially softened wording.

major comments (4)
  1. [Sections 3.1–3.7, Figures 6–12]
  2. [Sections 2.3 and 2.5, Abstract]
  3. [Section 3.6, Figure 11, Section 6, Section J]
  4. [Section 3.7, Figure 12, Section J]
minor comments (7)
  1. [Section 2.3] “EMS-Fold (Rives et al., 2019)” appears to be a citation error: Rives et al. is the ESM language-model paper, and the ESMFold structure-prediction tool should be cited with its own reference or the tool name corrected.
  2. [Figure 15 caption] “50 GDP-like protein backbones” should read “GFP-like”; GDP is the Ras ligand, not the GFP family.
  3. [Section 3.7] “Cytrochromec” is a typo for “Cytochrome c” in the heading and in the text following it.
  4. [Section 2.4] The sentence “We construct structural phylogenetic trees using 1−Q score as a distance measure, where higher values indicate greater structural similarity” is contradictory: if 1−Q is a distance, higher values indicate lower similarity.
  5. [Figure 7 caption] “HMC” is likely a typo for “HEC” (heme c); please make the ligand abbreviations consistent with Figure 12.
  6. [Sections 3.4 and 3.7 vs. Figure 8 caption] The PDB code for GDP-bound KRas is given as 4OBE in the text but as 4O8E in the Figure 8 caption; please unify the identifier.
  7. [Section 6] Claims such as “SM better captures conserved regions” and “FM offers greater flexibility” are made without quantitative uncertainty intervals or significance tests across the 50 samples per condition; consider adding error bars or statistical tests.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the pipeline's outputs are compared to family structures used in fine-tuning, but no claimed prediction reduces by construction to a fitted parameter or a self-citation chain.

full rationale

The paper's derivation chain is a generative pipeline (SE(3) score matching / flow matching) plus an evaluation protocol. The generation losses (Eqs. 6-7, 11-18) train models to approximate the family data distribution; the evaluation then asks whether sampled backbones are geometrically plausible (Ramachandran), conserve residues (ConSurf/Rate4Site), cluster by family (Qscore/3Di), remain stable in MD, and dock to family ligands. None of these is a fitted parameter renamed as a prediction, and no equation defines the validation metric in terms of the training loss. The MD and AutoDock Vina results are physics-based and independent of the learned scores, providing external (computational) grounding. The residual concern is that family identity and conserved-residue agreement are measured against the same experimentally derived family structures used for fine-tuning; the authors themselves note in Section 4 that 'there are regions of the observed diversity that are not so well covered (Figure 9) (such as GFP and class A beta-lactamases), raising concerns about potential overfitting and limited generalizability.' That is a correctable validation weakness, not a circular derivation under the strict criteria here, because the central functionality claims also rest on MD, docking, and structural-quality checks rather than solely on reproduction of the training distribution.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The paper introduces no free parameters fitted to data; the Qscore constant R0 equals 4 A and TM-score thresholds come from cited literature. The central claims rely on standard domain assumptions about geometric representation, self-consistency of sequence design, evolutionary rate computation, short MD simulations, and docking accuracy. No new physical entities are invented.

assumptions (5)
  • domain assumption Idealized backbone rigid groups (from AlphaFold) with fixed bond lengths and angles are a sufficient representation for generating realistic protein backbones.
    Adopted in Section B.2 from Yim et al. (2023); if the fixed-geometry assumption is too coarse, generated backbones could be geometrically implausible.
  • domain assumption ProteinMPNN-designed sequences that recover the generated backbone (high TMscore with ESMFold) are a valid proxy for designability.
    Section 2.3; this is a standard self-consistency proxy, but it is not experimentally verified and rests on the cited ESMFold model.
  • domain assumption Evolutionary rates computed by ConSurf from alignments of generated sequences with natural family sequences indicate functional conservation.
    Section 2.3; the alignments are against the same family data used for training, which is the source of the circularity concern.
  • domain assumption A 10 ns CHARMM36 explicit-solvent MD simulation is sufficient to infer dynamic stability under physiological conditions.
    Section 2.5 and I.1; short timescale, though consistent with common practice in the field.
  • domain assumption AutoDock Vina blind docking with a whole-protein grid box and lowest-energy pose selection can identify biologically relevant binding modes.
    Section 2.6; validated on experimental structures with ligand RMSD near 1.5 A, but remains an approximate scoring method.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Challenges and Guidelines in Deep Generative Protein Design: Four Case Studies." pith.science (2026). https://pith.science/paper/K7JXHWZY

@misc{pith2026241118568,
  author       = {Pith},
  title        = {Pith review of: Challenges and Guidelines in Deep Generative Protein Design: Four Case Studies},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/K7JXHWZY}},
  note         = {Machine review of arXiv:2411.18568}
}
abstract

Deep generative models show promise for $\textit{de novo}$ protein design, yet reliably producing designs that are geometrically plausible, evolutionarily consistent, functionally relevant, and dynamically stable remains challenging. We present a deep generative modeling pipeline for early $\textit{de novo}$ design of monomeric proteins, based on Score Matching and Flow Matching. We apply this pipeline to four diverse protein families with an adaptable evaluation protocol. Generated structures display realistic, clash-free conformations enriched with family-specific features, while the designed sequences preserve essential functional residues while retaining variability. Molecular dynamics and binding simulations show dynamic stability, with wild-type-like binding pockets that interact favorably with family-specific ligands. These results provide practical guidelines for integrating generative models into protein design workflows.

Figures

Figures reproduced from arXiv: 2411.18568 by the authors.

Figure 1
Figure 1. A schematic overview of the deep generative protein design pipeline and its evaluation protocols. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Protein backbone with dihedral angles ψ and ϕ. Molecules can be intuitively represented as 3D atomic point clouds. However, macromolecules like proteins may contain thousands or tens of thousands of atoms, with variation in the atom types and quantities among different amino acids (for instance, sulfur atoms are present only in a few amino acids like cysteine). Representing proteins as unordered 3D atomic point clou… view at source ↗
Figure 3
Figure 3. Overview of the (A) embedding module and (B) multi-layer network architecture. The networks s(θ, t, ·) involved in SM and v(θ, t, ·) involved in FM models, as reviewed in Section B, can share a common high-level architecture (Yim et al., 2023; Bose et al., 2023; Anand & Achim, 2022). 17 [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: (A) Overlay of WT E. coli TEM1 (PDB: 1BTL; Jelsch et al. (1993); green) and its E166N acylated intermediate (PDB: 1FQG; Brown et al. (2009); white) with penicillin (PNM). (B) WT E. caballus heart cytochrome c (PDB: 1HRC; (Dickerson et al., 1967)) with heme C (HEC). (C)…
Figure 5
Figure 5. Figure 5: Distributions of amino acid sequence lengths (aa) for the experimentally derived protein structures used for training (top row; orange) and the 50 backbone structures generated by each model (bottom row; blue). Target sequence lengths for the generated structures were …
Figure 6
Figure 6. Figure 6: Ramachandran plots comparing dihedral angles ψ and ϕ distributions for generated versus experimentally derived proteins; inset shows favored (light) and disallowed (dark) regions. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_6.png]
Figure 7
Figure 7. Figure 7: Normalized evolutionary rates mapped onto structures, with white for rapidly evolving positions and black for conserved ones. Experimentally derived structures are highlighted with bold outlines. Refer to [PITH_FULL_IMAGE:figures/full_fig_p022_7.png]
Figure 8
Figure 8. Figure 8: shows that, except for β-lactamase, the FM-generated sequences show a slightly higher average pairwise distance than those of the SM, indicating greater diversity. One possible explanation is that the experimentally derived β-lactamase structures used for training (fro…
Figure 9
Figure 9. Figure 9: Summary structural phylogenetic tree constructed using the Qscore and the 3Di alphabet. Branch colors distinguish families, with orange and blue nodes representing SM- and FM-generated structures, respectively. In β-lactamases, distinct Ambler classes are differentiate…
Figure 10
Figure 10. Figure 10: (A) Ultrametric summary structural phylogenetic tree constructed using the Qscore and the 3Di alphabet. (B) Ultrametric sequence-based phylogenetic tree constructed using the Clustal Omega and the FastTree pipeline. Different protein families are differentiated by dis…
Figure 11
Figure 11. Figure 11: Stability assessment of MD simulations across proteins using various metrics. Distributions of (A) RMSD, (B) radius of gyration (Rg), (C) potential energy, and (D) secondary structure counts throughout the simulation. Interquartile ranges and whiskers show metric vari…
Figure 12
Figure 12. Figure 12: Comparison of experimentally derived binding modes (bold boxes; first row) with predictions from blind docking simulations for receptors binding to family-specific ligands. Ligands are colored as green (ground truth), cyan (predicted using experimentally derived recep…
Figure 13
Figure 13. Figure 13: 50 β-lactamase-like protein backbones generated using score matching and 50 using flow matching. 32 [PITH_FULL_IMAGE:figures/full_fig_p032_13.png]
Figure 14
Figure 14. Figure 14: 50 cytochrome c-like protein backbones generated using score matching and 50 using flow matching. 33 [PITH_FULL_IMAGE:figures/full_fig_p033_14.png]
Figure 15
Figure 15. Figure 15: 50 GDP-like protein backbones generated using score matching and 50 using flow matching. 34 [PITH_FULL_IMAGE:figures/full_fig_p034_15.png]
Figure 16
Figure 16. Figure 16: 50 Ras-like protein backbones generated using score matching and 50 using flow matching. 35 [PITH_FULL_IMAGE:figures/full_fig_p035_16.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 11 canonical work pages

  1. [4]

    freeradbiomed.2009.03.004

    URL http://dx.doi.org/10.1016/j. freeradbiomed.2009.03.004. Kashyap, D., Garg, V . K., and Goel, N.Intrinsic and extrinsic pathways of apoptosis: Role in cancer de- velopment and prognosis, pp. 73–120. Elsevier, 2021. ISBN 9780323853156. doi: 10.1016/bs.apcsb.2021. 01.003. URL http://dx.doi.org/10.1016/bs. apcsb.2021.01.003. Kauke, M. J., Traxlmayr, M. W....

    doi:10.1016/j 2009

  2. [12]

    URL http: //dx.doi.org/10.1074/jbc.271.31.18379

    doi: 10.1074/jbc.271.31.18379. URL http: //dx.doi.org/10.1074/jbc.271.31.18379. Moi, D., Bernard, C., Steinegger, M., Nevers, Y ., Lan- gleib, M., and Dessimoz, C. Structural phylogenetics unravels the evolutionary diversification of communica- tion systems in gram-positive bacteria and their viruses. September 2023. doi: 10.1101/2023.09.19.558401. 8 Chal...

    work page doi:10.1074/jbc.271.31.18379

  3. [21]

    doi: 10.1002/jcc.21334

    ISSN 1096-987X. doi: 10.1002/jcc.21334. URL http://dx.doi.org/10.1002/jcc.21334. van Kempen, M., Kim, S. S., Tumescheit, C., Mirdita, M., Lee, J., Gilchrist, C. L. M., S ¨oding, J., and Steinegger, M. Fast and accurate protein structure search with foldseek.Nature Biotechnology, 42(2): 243–246, May 2023. ISSN 1546-1696. doi: 10.1038/ s41587-023-01773-0. U...

    doi:10.1002/jcc.21334

  4. [23]

    face-lifted

    ISSN 0959-440X. doi: 10.1016/j.sbi.2024. 102794. URL http://dx.doi.org/10.1016/j. sbi.2024.102794. Wong, H. Y . and Wong, K.-B.Using AlphaFold2 and Molec- ular Dynamics Simulation to Model Protein Recognition, pp. 49–66. Springer US, 2024. ISBN 9781071640593. doi: 10.1007/978-1-0716-4059-3 4. URL http://dx. doi.org/10.1007/978-1-0716-4059-3_4. Wu, K. E., ...

    arXiv 2024

  5. [25]

    URL https: //arxiv.org/abs/2302.02277

    doi: 10.48550/ARXIV .2302.02277. URL https: //arxiv.org/abs/2302.02277. Yu, W., He, X., Vanommeslaeghe, K., and MacKerell, A. D. Extension of the charmm general force field to sulfonyl- containing compounds and its utility in biomolecular simulations.Journal of Computational Chemistry, 33 (31):2451–2468, July 2012. ISSN 1096-987X. doi: 10.1002/jcc.23067. ...

    arXiv doi:10.48550/arxiv.2302.02277

  6. [370]

    URL http:// dx.doi.org/10.1016/j.csbj.2019.12.004

    doi: 10.1016/j.csbj.2019.12.004. URL http:// dx.doi.org/10.1016/j.csbj.2019.12.004. Patient, S., Wieser, D., Kleen, M., Kretschmann, E., Je- sus Martin, M., and Apweiler, R. Uniprotjapi: a remote api for accessing uniprot data.Bioinformat- ics, 24(10):1321–1322, April 2008. ISSN 1367-4803. doi: 10.1093/bioinformatics/btn122. URL http://dx. doi.org/10.1093...

    arXiv 2019

  7. [896]

    URL http:// dx.doi.org/10.1016/j.bmc.2016.06.034

    doi: 10.1016/j.bmc.2016.06.034. URL http:// dx.doi.org/10.1016/j.bmc.2016.06.034. Tan, G., Gil, M., L¨oytynoja, A. P., Goldman, N., and Dessi- moz, C. Simple chained guide trees give poorer multiple sequence alignments than inferred trees in simulation and phylogenetic benchmarks.Proceedings of the Na- tional Academy of Sciences, 112(2), January 2015. ISS...

    arXiv 2016

  8. [1719]

    URL http:// dx.doi.org/10.1093/molbev/msaa100

    doi: 10.1093/molbev/msaa100. URL http:// dx.doi.org/10.1093/molbev/msaa100. Mayrose, I. Comparison of site-specific rate-inference methods for protein sequences: Empirical bayesian methods are superior.Molecular Biology and Evolu- tion, 21(9):1781–1791, May 2004. ISSN 1537-1719. doi: 10.1093/molbev/msh194. URL http://dx.doi. org/10.1093/molbev/msh194. McI...

    work page doi:10.1093/molbev/msaa100

Show all 26 references
  1. [1963]

    doi: 10.1016/s0022-2836(63) 80023-6

    ISSN 0022-2836. doi: 10.1016/s0022-2836(63) 80023-6. URL http://dx.doi.org/10.1016/ S0022-2836(63)80023-6. Remington, S. J. Green fluorescent protein: A perspective. Protein Science, 20(9):1509–1519, July 2011. ISSN 1469- 896X. doi: 10.1002/pro.684. URL http://dx.doi. org/10.1...

    2011 doi

  2. [1967]

    doi: 10.1016/0022-2836(67) 90182-9

    ISSN 0022-2836. doi: 10.1016/0022-2836(67) 90182-9. URL http://dx.doi.org/10.1016/ 0022-2836(67)90182-9. Du, Y ., Jamasb, A. R., Guo, J., Fu, T., Harris, C., Wang, Y ., Duan, C., Li `o, P., Schwaller, P., and Blundell, T. L. Machine learning-aided generative molecular design.N...

    doi

  3. [1977]

    doi: 10.1146/annurev.bi.46

    ISSN 1545-4509. doi: 10.1146/annurev.bi.46. 070177.001503. URL http://dx.doi.org/10. 1146/annurev.bi.46.070177.001503. Sievers, F., Wilm, A., Dineen, D., Gibson, T. J., Karplus, K., Li, W., Lopez, R., McWilliam, H., Remmert, M., S¨oding, J., Thompson, J. D., and Higgins, D. G....

  4. [1996]

    doi: 10.1007/bf00228148

    ISSN 1573-5001. doi: 10.1007/bf00228148. URL http://dx.doi.org/10.1007/BF00228148. Laskowski, R. A., MacArthur, M. W., Moss, D. S., and Thornton, J. M. Procheck: a program to check the stereochemical quality of protein struc- tures.Journal of Applied Crystallography, 26(2): 28...

    doi

  5. [2001]

    doi: 10.1002/ijc.1322

    ISSN 1097-0215. doi: 10.1002/ijc.1322. URL http://dx.doi.org/10.1002/ijc.1322. Lipman, Y ., Chen, R. T. Q., Ben-Hamu, H., Nickel, M., and Le, M. Flow matching for generative modeling, 2022. URLhttps://arxiv.org/abs/2210.02747. Liu, K., Watanabe, E., and Kokubo, H. Exploring th...

    2022 arXiv

  6. [2002]

    doi: 10.1093/bioinformatics/ 18.suppl\ 1.s71

    ISSN 1367-4803. doi: 10.1093/bioinformatics/ 18.suppl\ 1.s71. URL http://dx.doi.org/10. 1093/bioinformatics/18.suppl_1.S71. Pushparathinam, G. and Kathiravan, M. K. Docking studies and molecular dynamics simulation of triazole benzene sulfonamide derivatives with human carboni...

    doi

  7. [2008]

    doi: 10.1038/nrm2434

    ISSN 1471-0080. doi: 10.1038/nrm2434. URL http://dx.doi.org/10.1038/nrm2434. O’Boyle, N. M., Banck, M., James, C. A., Morley, C., Vandermeersch, T., and Hutchison, G. R. Open ba- bel: An open chemical toolbox.Journal of Chem- informatics, 3(1), October 2011. ISSN 1758-2946. do...

    doi

  8. [2009]

    doi: 10.1002/prot.22458

    ISSN 1097-0134. doi: 10.1002/prot.22458. URL http://dx.doi.org/10.1002/prot.22458. Ingraham, J. B., Baranov, M., Costello, Z., Barber, K. W., Wang, W., Ismail, A., Frappier, V ., Lord, D. M., Ng-Thow-Hing, C., Van Vlack, E. R., Tie, S., Xue, V ., Cowles, S. C., Leung, A., Rodr...

    arXiv

  9. [2012]

    doi: 10.1021/ci300363c

    ISSN 1549-960X. doi: 10.1021/ci300363c. URL http://dx.doi.org/10.1021/ci300363c. 10 Challenges and Guidelines in Deep Generative Protein Design: Four Case Studies Vanommeslaeghe, K., Hatcher, E., Acharya, C., Kundu, S., Zhong, S., Shim, J., Darian, E., Guvench, O., Lopes, P., ...

    arXiv

  10. [2018]

    doi: 10.1016/j.neuron.2018

    ISSN 0896-6273. doi: 10.1016/j.neuron.2018. 08.011. URL http://dx.doi.org/10.1016/j. neuron.2018.08.011. 6 Challenges and Guidelines in Deep Generative Protein Design: Four Case Studies Hooft, R. W. W., Vriend, G., Sander, C., and Abola, E. E. Errors in protein structures.Natu...

    2018 doi

  11. [2019]

    URL https://www

    doi: 10.1101/622803. URL https://www. biorxiv.org/content/10.1101/622803v4. Rudolph, M., Wandt, B., and Rosenhahn, B. Same same but differnet: Semi-supervised defect detection with nor- malizing flows, 2020. URL https://arxiv.org/ abs/2008.12577. Salemme, F. R. Structure and f...

    2020 arXiv

  12. [2022]

    doi: 10.1126/science.add2187

    ISSN 1095-9203. doi: 10.1126/science.add2187. URL http://dx.doi.org/10.1126/science. add2187. De Bortoli, V ., Mathieu, E., Hutchinson, M., Thornton, J., Teh, Y . W., and Doucet, A. Riemannian score-based generative modelling, 2022. URL https://arxiv. org/abs/2202.02763. Dicke...

    arXiv

  13. [2023]

    doi: 10.1002/pro.4582

    ISSN 1469-896X. doi: 10.1002/pro.4582. URL http://dx.doi.org/10.1002/pro.4582. Yim, J., Trippe, B. L., De Bortoli, V ., Mathieu, E., Doucet, A., Barzilay, R., and Jaakkola, T. SE(3) diffusion model with application to protein backbone generation.ICML,

    doi

  14. [2024]

    URL https://doi.org/10.21105/joss. 06943. Morris, G. M., Huey, R., Lindstrom, W., Sanner, M. F., Belew, R. K., Goodsell, D. S., and Olson, A. J. Autodock4 and autodocktools4: Automated docking with selective receptor flexibility.Journal of Computational Chemistry, 30(16):2785–...

  15. [2836]

    attritional

    doi: 10.1006/jmbi.1993.1626. URL http://dx. doi.org/10.1006/jmbi.1993.1626. 12 Challenges and Guidelines in Deep Generative Protein Design: Four Case Studies A. Deep Generative Protein Design Workflow Figure 1.A schematic overview of the deep generative protein design pipeline...

    2023

  16. [4951]

    URL http:// dx.doi.org/10.1007/s10822-016-0005-2

    doi: 10.1007/s10822-016-0005-2. URL http:// dx.doi.org/10.1007/s10822-016-0005-2. Malik, A. J., Poole, A. M., and Allison, J. R. Structural phylogenetics with confidence.Molecular Biology and Evolution, 37(9):2711–2726, April 2020. ISSN 1537-

    doi

  17. [4962]

    URL http://dx

    doi: 10.1093/nar/28.1.235. URL http://dx. doi.org/10.1093/nar/28.1.235. Bose, A. J., Akhound-Sadegh, T., Huguet, G., Fatras, K., Rector-Brooks, J., Liu, C.-H., Nica, A. C., Korablyov, M., Bronstein, M., and Tong, A. SE(3)-stochastic flow matching for protein backbone generatio...

    arXiv

  18. [9258]

    URL http: //dx.doi.org/10.1074/jbc.M109.053819

    doi: 10.1074/jbc.m109.053819. URL http: //dx.doi.org/10.1074/jbc.M109.053819. Burton, B., Zimmermann, M. T., Jernigan, R. L., and Wang, Y . A computational investigation on the con- 5 Challenges and Guidelines in Deep Generative Protein Design: Four Case Studies nection betwee...

    doi arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.