Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

Inverse problems with experiment-guided AlphaFold

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper claims that AlphaFold3 can be used as a structural prior and steered by experimental measurements to generate protein ensembles that fit the data, sometimes better than deposited structures.

desk verdict Solid demonstration that guided AF3 ensembles fit density and NOE data, but the guidance effect is confounded with base model choice and the evaluation is in-sample. read the letter →

arxiv 2502.09372 v2 pith:GMYEG26K submitted 2025-02-13 q-bio.BM

Advaith Maddipatla , Nadav Bojan Sellam , Meital Bojan , Sanketh Vedula , Paul Schanda , Ailie Marx , Alex M. Bronstein This is my paper · ORCID
classification q-bio.BM
keywords proteinstructureensemblesdiffusionguidanceinverseproblemsexperimentaldataconditioningX-raycrystallographyNMRrestraintsAlphaFold3conformationalheterogeneity
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

The paper tries to show that a state-of-the-art structure predictor can be turned into a generator of experimentally consistent conformational ensembles instead of a single static model. The authors treat AlphaFold3's diffusion module as a sequence-conditioned prior over protein structures and use Bayes' rule to sample the posterior given experimental measurements: electron density maps, NMR distance restraints, or known atom positions. They build a three-stage pipeline — guided diffusion, force-field relaxation, and greedy ensemble selection — and test it on crystallographic and NMR data. If the central claim holds, structural biology gains a way to produce physically plausible, data-consistent ensembles in minutes, including conformational heterogeneity that single-structure prediction misses.

What carries the argument

The machinery is a non-i.i.d. guided diffusion sampler built on AlphaFold3's variance-preserving stochastic differential equation. At each diffusion step the sampler evolves $n$ structures jointly, adding the gradient of the experimental log-likelihood — evaluated on ensemble averages rather than on individual members — to the unconditional score. The likelihoods are an $\ell^1$ distance between observed and calculated electron density, a piecewise quadratic penalty on ensemble-averaged NOE distances, and a quadratic substructure anchor. The resulting samples are filtered for clashes and broken bonds, relaxed with a harmonic force field, and pruned by matching pursuit to a small set that maximizes the likelihood.

What would settle it

Hold out a random half of the ubiquitin NOE restraints, run NOE-guided AlphaFold3 using only the other half, and compare the resulting ensemble against the held-out restraints and the experimental N-H $S^2$ order parameters; if the guided ensemble is no better than an unguided AlphaFold3 ensemble of the same size on held-out data, the guidance is fitting the input restraints rather than recovering the posterior.

Watch

Extended reading notes

Core claim

The central claim is that posterior inference over protein structures, with AlphaFold3's diffusion score as the prior and a differentiable forward model of an experiment as the likelihood, yields ensembles faithful to that experiment. Concretely, the paper reports that density-guided AlphaFold3 recovers alternate backbone conformations in crystal structures that unguided AlphaFold3 predicts as a single mode, and that NOE-guided AlphaFold3 produces ubiquitin ensembles whose N-H $S^2$ order parameters correlate with experiment at $r=0.72$, versus $r=0.47$ for unguided AlphaFold3, while also reducing restraint violations. In several cases the guided ensembles fit the experimental data better than structures deposited in the public archive. The authors frame this as a general inverse-problem formulation: any experimental modality with a computable likelihood can be plugged into the same guided-sampling procedure.

Load-bearing premise

The load-bearing premise is that AlphaFold3's diffusion score is a valid prior for posterior inference over ensembles despite being trained on static crystallographic models, and that the simple forward models — a uniform B-factor for density and heavy-atom substitution for NOE distances — are accurate enough that guidance improves rather than corrupts the posterior.

Editorial extensions

If this is right

  • Crystallographic model building could recover alternate-conformation and flexible-region heterogeneity in minutes, starting from a sequence and a density map, without manual multi-conformer refinement.
  • NMR ensemble determination could move from days of restrained molecular dynamics to minutes of guided sampling, with equal or better restraint compliance.
  • The inverse-problem formulation is modular: any experimental observable with a differentiable forward model can be added as a likelihood, so the same sampler should extend to cryo-EM, small-angle scattering, or chemical shifts without retraining.
  • Guided ensembles can serve as a quantitative cross-check on deposited structures, because in some reported cases they fit the experimental data better than the PDB entry.

Reading between the lines

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

  • Extension: the same non-i.i.d. ensemble likelihood could be applied to data that constrain only a subset of atoms, such as DEER distance distributions between spin labels, by replacing the substructure anchor with a likelihood over label positions; the paper does not test this.
  • Testable extension: because the density forward model uses a single B-factor $B=4/n$ for all atoms, replacing it with per-atom or resolution-dependent B-factors after sampling is a natural way to separate forward-model error from prior error.
  • Testable extension: the NOE likelihood averages distances linearly, whereas the measured signal averages $r^{-6}$; switching to an $r^{-6}$ ensemble average could change which ensembles satisfy the data, especially for mobile residues.
  • If the diffusion prior is the main source of physical plausibility, the same guidance recipe should port to any future diffusion-based structure predictor, so improvements in the base model would transfer without changing the inverse-problem layer.
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 / 4 minor

Summary. The paper proposes an "experiment-guided AlphaFold3" framework: using a diffusion-based structure predictor (the open-source Protenix re-implementation of AlphaFold3) as a prior, it samples non-i.i.d. structural ensembles under guidance from crystallographic electron density maps, NOE distance restraints, and substructure conditioning, followed by force-field relaxation and a matching-pursuit ensemble selection step. The authors demonstrate the approach on crystallographic proteins with alternative conformations and on NMR proteins with NOE restraints, reporting improved density cosine similarity, reduced NOE violations, better capture of bimodal backbone conformations, and improved agreement with independent N-H S2 order parameters for ubiquitin. The central claim is that experiment-guided AlphaFold3 produces ensembles consistent with measured data and outperforms unguided AlphaFold3, sometimes fitting experimental data better than PDB-deposited structures.

Significance. If the central claim holds, this is a useful contribution: it casts ensemble modeling as posterior inference under a diffusion prior, shows that a generic inverse-problem formulation can accommodate several experimental modalities, and demonstrates substantial runtime gains over traditional restrained-MD pipelines. The framework is general, the real-data evaluation is broad (multiple crystallographic targets, 100-NMR-spectra benchmark entries, and an expanded NMRDb set), and the ubiquitin S2 validation is a genuinely independent check. The method's open-source implementation basis (Protenix) also aids reproducibility. However, the empirical case is currently weakened by two structural issues: the guided method is not compared against its own unguided base model, and most reported agreement metrics are computed on the same experimental observations used for guidance and ensemble selection.

major comments (4)
  1. [A.5 and Tables A1–A4, A7–A10] The headline comparison is guided Protenix versus official AlphaFold3, not guided versus unguided Protenix. Appendix A.5 states that all experiments use Protenix, while the AlphaFold3 baseline is generated with official AlphaFold3 weights. Tables A1–A4, A7–A10, and the ubiquitin validation in Figure 5 therefore conflate two differences: the presence of experimental guidance and the choice of base model. If Protenix and official AlphaFold3 differ in diffusion schedule, output diversity, or calibration, the observed gains may be caused by the base model alone. Please add an unguided Protenix baseline processed through the same sampling, filtering, relaxation, and (where applicable) matching-pursuit pipeline, and report it in every main comparison table.
  2. [3.1, 4.3, and Tables A1–A3, A9] The crystallographic evaluation is in-sample with respect to the reported metric. Density guidance in equation (1), the post-hoc B-factor optimization in A.8, and the matching-pursuit selection in Algorithm 2 all maximize agreement between the model and the same observed map Fo whose cosine similarity is then reported in Tables A1–A3 and A9. The claim that guided ensembles 'fit the density better than the PDB structure' is consequently a statement about fitting the training objective, not an independent validation of the method. Please provide an R-free-style or withheld-reflection evaluation, or an omit-map analysis, so that the density agreement is assessed on data not used for guidance and selection.
  3. [6, Eq. (2), and A.1] The NOE violation percentages and violation distances in Table A4 are computed against the same restraints used to guide sampling, and the guidance strength eta is selected in A.1 to maximize restraint compliance on that same dataset. These numbers therefore describe goodness-of-fit rather than predictive accuracy. The strength of the NMR section is the independent S2 comparison in Figure 5; please state explicitly in the main text that the restraint-based metrics are in-sample fit metrics, and consider reporting cross-validation on a held-out subset of restraints or using an independent observable for all proteins, not only ubiquitin.
  4. [A.12] The heavy-atom approximation used during NOE guidance introduces a stated maximum error of up to 4.4 Å, while evaluation in A.12 uses explicit hydrogen positions. The paper acknowledges this discrepancy, but its effect on the shape of the guided posterior is not assessed. A practical check would be to compare ensembles guided with the current approximation against ensembles guided with hydrogens placed at the relaxation stage, at least on a few benchmark systems, to confirm that the approximation is not a dominant error source.
minor comments (4)
  1. [A.5 and table captions] The term 'AlphaFold3' is used both for the official baseline and as part of 'Guided AlphaFold3' (which uses Protenix). Please state this distinction directly in every table caption and figure legend, since readers may otherwise assume a single base model.
  2. [A.11] The text contains a typo: 'shapr reduction' should be 'sharp reduction'.
  3. [1 and 5] Minor wording issues: 'these approaches remainstrained' in Section 1 should read 'remain trained', and 'an other 11 cases' in Section 5 should be 'another 11 cases'.
  4. [Abstract and A.4] The abstract and Section 6 claim the method is 'orders of magnitude faster than the status quo', but Tables A5–A6 compare runtime only against unguided AlphaFold3. A direct runtime comparison with a restrained-MD or CYANA pipeline would substantiate this claim.

Circularity Check

2 steps flagged · score 6.0 of 10

Density and NOE headline results are evaluated against the same experimental observations that the pipeline explicitly optimizes; the independent S2 and bimodality checks carry the external validity.

  1. fitted input called prediction [Sections 4.3 and A.8; evaluated in Tables A1–A3, A9]
    "Post relaxation, we employ a matching pursuit-based approach (Mallat & Zhang, 1993) to greedily select a subset of the relaxed ensemble, X I ={X k :k∈ I} , that best fits the observation y. ... we seek to maximize logp(y|X I∪{k} ,a) over all k /∈ I. ... Before selecting the ensemble, we optimize a scalar B-factor B to maximize the log-likelihood in equation (1)."

    The reported density metrics (cosine similarity between observed Fo and calculated Fc in Tables A1–A3 and A9) measure agreement with the same Fo used in the likelihood that matching pursuit explicitly maximizes, and the uniform B-factor is post-hoc tuned against that same Fo. The 'better fit than PDB/AlphaFold3' numbers are therefore optimized fitting scores, not out-of-sample predictions; the claim is partly true by construction, though the recovered altloc geometry and bimodality distributions provide independent evidence.

  2. fitted input called prediction [Sections 3.2 and 6; evaluated in Tables A4, A7, A10]
    "With some abuse of NMR physics, we henceforth assume that the distance average is observed directly, and define the NOE constraints as a set D of pairs of lower and upper bounds on the ensemble average ... The log-likelihood is given by logp(D|X,a)=... In all cases, NOE-guided AlphaFold reduced violations, and in half of the cases, the guided ensembles even outperformed PDB-deposited NMR structures in restraint compliance."

    Violation percentages and violation distances in Table A4 are computed on exactly the same restraint set D used as the guidance objective in equation (2). The claim that guided ensembles 'adhere to experimental NOE data more faithfully' and 'outperform PDB' in restraint compliance is thus an optimization-success report on the training target, not a test on withheld NOEs. The independent S2 order-parameter correlation in Figure 5 uses a different observable and is the genuinely external validation.

full rationale

The central density and NMR results are partially circular because the evaluation metrics are computed against the same experimental observations used in the optimization: matching pursuit and B-factor tuning maximize the density likelihood before the density cosine similarities are reported, and NOE guidance minimizes violations of restraints that are then counted in the violation tables. These are not completely vacuous successes, since the AlphaFold prior can resist guidance and the L1 density objective is not identical to the cosine evaluation, but presenting them as 'outperforming PDB/AlphaFold3' mixes fitting behavior with discovery. The external load is carried mainly by the S2 validation, the bimodality analysis against PDB altloc coordinates, and the visual structural checks. The Protenix-versus-official-AlphaFold3 baseline mismatch (Appendix A.5) is a real confound for attribution of gains to guidance, but it is a correctness risk rather than circularity. The paper's own Appendix A.12 acknowledges the heavy-atom approximation (up to 4.4 Å error) and the approximate intensity-to-distance averaging, which weaken the forward model but do not make the derivation circular. The altloc benchmark from Rosenberg et al. (2024a) is a self-citation with overlapping authors, but it is an externally checkable PDB-derived dataset, so it is not load-bearing circularity. Overall, two headline 'prediction' families reduce substantially to their own fitting objectives, meriting a score of 6.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

No new physical entities are introduced. The framework relies on standard diffusion guidance and adds hand-chosen hyperparameters and simplified forward models; the B-factor, guidance scale, and region selection are fitted to the experimental data used for evaluation.

free parameters (6)
  • B-factor B = optimized per protein (not reported)
    Used as bandwidth in the calculated electron density Fc; set to B=4/n during guidance and then optimized with Adam over 100 iterations against Fo before ensemble selection (A.6, A.8).
  • Guidance scale eta = 0.1 (density), 0.3/0.5 (NOE)
    Scales the guidance score in Eq. 5; selected manually based on restraint compliance for NOE (A.1).
  • Substructure conditioner scale lambda = 0.1
    Weights the anchoring penalty for atoms outside the refined region in density guidance (A.1).
  • Ensemble size n = 16 (density), varies (NOE)
    Number of structures jointly sampled; chosen for out-of-memory constraints and used in all density experiments (A.6).
  • Maximum ensemble size nmax = 5
    Matching-pursuit stopping criterion, heuristically chosen so the density is well explained without overfitting noise (A.8).
  • Bond/clash thresholds = tau_bond=2.1 Å, tau_clash=1.1 Å
    Hand-set filters for broken bonds and steric clashes (A.7).
assumptions (6)
  • domain assumption AlphaFold3/Protenix diffusion model provides a valid prior p(X|a) for protein structure ensembles.
    Invoked in Section 3 and Eq. 5; the paper itself notes AlphaFold3 is trained on static X-ray models, so the prior may be biased toward rigid states.
  • domain assumption The guided diffusion SDE in Eq. 5 samples from the posterior p(X|a,y).
    No proof is given that the modified SDE with added guidance score converges to the posterior; standard heuristic in diffusion guidance.
  • domain assumption The forward model for electron density using a uniform B-factor and tabulated form factors is accurate enough for guidance.
    Fc is computed via KDE with B=4/n, and the B-factor is optimized post hoc; this is a simplified noise model (A.2, A.6).
  • domain assumption NOE restraints can be modeled as bounds on ensemble-averaged distances.
    The paper approximates NOE intensities by distance averages, acknowledging the rigorous r^-6 intensity averaging is left for future work (A.12).
  • domain assumption Heavy-atom substitution for hydrogen in NOE restraints introduces negligible error.
    A.12 states maximum error is 4.4 Å, which is not negligible for restraints with 5-6 Å upper bounds.
  • ad hoc to paper The residue region r to refine is known in advance.
    All density-guided experiments operate on pre-specified residue regions, often taken from known altloc annotations (Tables A1-A3, A9). The method does not discover these regions from density alone.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Inverse problems with experiment-guided AlphaFold." pith.science (2026). https://pith.science/paper/GMYEG26K

@misc{pith2026250209372,
  author       = {Pith},
  title        = {Pith review of: Inverse problems with experiment-guided AlphaFold},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GMYEG26K}},
  note         = {Machine review of arXiv:2502.09372}
}
read the original abstract

Proteins exist as a dynamic ensemble of multiple conformations, and these motions are often crucial for their functions. However, current structure prediction methods predominantly yield a single conformation, overlooking the conformational heterogeneity revealed by diverse experimental modalities. Here, we present a framework for building experiment-grounded protein structure generative models that infer conformational ensembles consistent with measured experimental data. The key idea is to treat state-of-the-art protein structure predictors (e.g., AlphaFold3) as sequence-conditioned structural priors, and cast ensemble modeling as posterior inference of protein structures given experimental measurements. Through extensive real-data experiments, we demonstrate the generality of our method to incorporate a variety of experimental measurements. In particular, our framework uncovers previously unmodeled conformational heterogeneity from crystallographic densities, and generates high-accuracy NMR ensembles orders of magnitude faster than the status quo. Notably, we demonstrate that our ensembles outperform AlphaFold3 and sometimes better fit experimental data than publicly deposited structures to the Protein Data Bank (PDB). We believe that this approach will unlock building predictive models that fully embrace experimentally observed conformational diversity.

Figures

Figures reproduced from arXiv: 2502.09372 by the authors.

Figure 1
Figure 1. Schematic depiction of the proposed method. AlphaFold3 allows the sampling of protein structures given an amino acid sequence. To further condition the model by experimental observations, at each time step of the AlphaFold3 diffusion model, an ensemble of structures is generated. Likelihoods of experimental observations are calculated given each individual ensemble member (e.g., to enforce a substructure) and on ens… view at source ↗
Figure 2
Figure 2. Two crystallographic observations of the SARS-CoV-2 ORF8 protein at 1.62A resolution (PDB: ˚ 7JX6 and 7F5F color coded as purple and green, respectively). The two structures exhibit major differences at three sites despite sharing a nearly identical main acid sequence deferring by a single mutation L67S (A). AlphaFold3 prediction appears practically identical for the two sequences mispredicting the structure of the … view at source ↗
Figure 3
Figure 3. Crystallographic observation of the human NBR1 protein at 2.52A resolution (PDB: ˚ 4OLE) exhibits a multi-modal backbone distribution at 423-431 (conformation modes A and B color-coded in green and purple, respectively). AlphaFold3 predicts only conformation B while completely missing the helical conformation A. Electron density-guided AlphaFold3 predicts a bi-modally distributed ensemble better describing the obser… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: NOE constraint violations in the Methanocaldococcus jannaschii MJ1198 protein in the NMR structure ensemble (PDB: 2K52) and ensembles predicted by AlphaFold3 and using NOE-guidance with strength 0.2 and 0.5. Violated constraints are depicted as lines color-coded by the…
Figure 5
Figure 5. Figure 5: Comparison of conformational flexibility in ubiquitin ensembles predicted by AlphaFold3 and our i.i.d and non-i.i.d NOE￾guided method. The N-H order parameter S 2 is calculated on the ensembles and compared to the experimental NMR observables. For guided predictions, p…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. 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.

  2. Solving Inverse Problems via Diffusion-Based Priors: An Approximation-Free Ensemble Sampling Approach

    cs.LG 2025-06 conditional novelty 5.0 of 10

    A weighted-particle sampler evolves the posterior through the diffusion model's reverse dynamics, with theoretical error bounds and improved image reconstructions.

Reference graph

Works this paper leans on

61 extracted references · 43 canonical work pages · cited by 2 Pith papers

  1. [1]

    J., Bambrick, J., et al

    Abramson, J., Adler, J., Dunger, J., Evans, R., Green, T., Pritzel, A., Ronneberger, O., Willmore, L., Ballard, A. J., Bambrick, J., et al. Accurate structure prediction of biomolecular interactions with alphafold 3. Nature, pp.\ 1--3, 2024

    2024

  2. [2]

    D., Afonine, P

    Adams, P. D., Afonine, P. V., Bunk \'o czi, G., Chen, V. B., Davis, I. W., Echols, N., Headd, J. J., Hung, L.-W., Kapral, G. J., Grosse-Kunstleve, R. W., et al. Phenix: a comprehensive python-based system for macromolecular structure solution. Acta Crystallographica Section D: Biological Crystallography, 66 0 (2): 0 213--221, 2010

    work page 2010

  3. [3]

    R., Wang, J., Cong, Q., Kinch, L

    Baek, M., DiMaio, F., Anishchenko, I., Dauparas, J., Ovchinnikov, S., Lee, G. R., Wang, J., Cong, Q., Kinch, L. N., Schaeffer, R. D., et al. Accurate prediction of protein structures and interactions using a three-track neural network. Science, 373 0 (6557): 0 871--876, 2021

    2021

  4. [4]

    Accurate prediction of protein--nucleic acid complexes using rosettafoldna

    Baek, M., McHugh, R., Anishchenko, I., Jiang, H., Baker, D., and DiMaio, F. Accurate prediction of protein--nucleic acid complexes using rosettafoldna. Nature methods, 21 0 (1): 0 117--121, 2024

    work page 2024

  5. [5]

    and Patiny, L

    Banfi, D. and Patiny, L. www. nmrdb. org: Resurrecting and processing nmr spectra on-line. Chimia, 62 0 (4): 0 280--280, 2008

    work page 2008

  6. [6]

    P., Hahn, T., Wondratschek, H., M \"u ller, U., Shmueli, U., Prince, E., Authier, A., Kopsk \`y , V., Litvin, D., Arnold, E., et al

    Brock, C. P., Hahn, T., Wondratschek, H., M \"u ller, U., Shmueli, U., Prince, E., Authier, A., Kopsk \`y , V., Litvin, D., Arnold, E., et al. International tables for crystallography volume A: Space-group symmetry. Wiley Online Library, 2016

    work page 2016

  7. [7]

    K., Berman, H

    Burley, S. K., Berman, H. M., Kleywegt, G. J., Markley, J. L., Nakamura, H., and Velankar, S. Protein data bank (pdb): the single global macromolecular structure archive. Protein crystallography: methods and protocols, pp.\ 627--641, 2017

    work page 2017

  8. [8]

    Protenix - advancing structure prediction through a comprehensive alphafold3 reproduction

    Chen, X., Zhang, Y., Lu, C., Ma, W., Guan, J., Gong, C., Yang, J., Zhang, H., Zhang, K., Wu, S., Zhou, K., Yang, Y., Liu, Z., Wang, L., Shi, B., Shi, S., and Xiao, W. Protenix - advancing structure prediction through a comprehensive alphafold3 reproduction. bioRxiv, 2025. doi:10.1101/2025.01.08.631967

    doi:10.1101/2025.01.08.631967 2025

Show all 61 references
  1. [9]

    DeLano, W. L. et al. Pymol: An open-source molecular graphics tool. CCP4 Newsl. Protein Crystallogr, 40 0 (1): 0 82--92, 2002

    2002

  2. [10]

    Diederik, P. K. Adam: A method for stochastic optimization. International Conference on Learning Representations, 2015

    2015

  3. [11]

    Doudna, J. A. Structural genomics of rna. Nature Structural Biology, 7 0 (11): 0 954--956, Nov 2000. ISSN 1545-9985

    2000

  4. [12]

    J., Terwilliger, T

    Fadini, A., Li, M., McCoy, A. J., Terwilliger, T. C., Read, R. J., Hekstra, D., and AlQuraishi, M. Alphafold as a prior: Experimental structure determination conditioned on a pretrained neural network. bioRxiv, 2025. doi:10.1101/2025.02.18.638828. URL https://www.biorxiv.org/c...

    2025 doi

  5. [13]

    and Argos, P

    Frishman, D. and Argos, P. Knowledge-based protein secondary structure assignment. Proteins: Structure, Function, and Bioinformatics, 23 0 (4): 0 566--579, 1995

    1995

  6. [14]

    L., DePristo, M

    Furnham, N., Blundell, T. L., DePristo, M. a., and Terwilliger, T. C. Is one solution good enough? Nat. Struct. Mol. Biol., 13 0 (3): 0 184--185; discussion 185, 2006

    2006

  7. [15]

    Correlated mutations and residue contacts in proteins

    G \"o bel, U., Sander, C., Schneider, R., and Valencia, A. Correlated mutations and residue contacts in proteins. Proteins: Structure, Function, and Bioinformatics, 18 0 (4): 0 309--317, 1994. doi:doi: 10.1002/prot.340180402

    1994 doi

  8. [16]

    Automated nmr structure calculation with cyana

    G \"u ntert, P. Automated nmr structure calculation with cyana. Protein NMR techniques, pp.\ 353--378, 2004

    2004

  9. [17]

    Modeling with alternate locations in x-ray protein structures

    Gutermuth, T., Sieg, J., Stohn, T., and Rarey, M. Modeling with alternate locations in x-ray protein structures. Journal of Chemical Information and Modeling, 63 0 (8): 0 2573--2585, 2023

    2023

  10. [18]

    Denoising diffusion probabilistic models

    Ho, J., Jain, A., and Abbeel, P. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33: 0 6840--6851, 2020

    2020

  11. [19]

    A., Sch \"a rfe, C

    Hopf, T. A., Sch \"a rfe, C. P., Rodrigues, J. P., Green, A. G., Kohlbacher, O., Sander, C., Bonvin, A. M., and Marks, D. S. Sequence co-evolution gives 3d contacts and structures of protein complexes. elife, 3: 0 e03430, 2014

    2014

  12. [20]

    Comparison of multiple amber force fields and development of improved protein backbone parameters

    Hornak, V., Abel, R., Okur, A., Strockbine, B., Roitberg, A., and Simmerling, C. Comparison of multiple amber force fields and development of improved protein backbone parameters. Proteins: Structure, Function, and Bioinformatics, 65 0 (3): 0 712--725, 2006

    2006

  13. [21]

    B., Baranov, M., Costello, Z., Barber, K

    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., Rodrigues, J. a. V., Morales-Perez, C. L., Ayoub, A. M., Green, R., Puentes, K., Oplinger...

    2023 doi

  14. [22]

    Alphafold meets flow matching for generating protein ensembles

    Jing, B., Berger, B., and Jaakkola, T. Alphafold meets flow matching for generating protein ensembles. In Forty-first International Conference on Machine Learning, 2024

    2024

  15. [23]

    Highly accurate protein structure prediction with alphafold

    Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Z \' dek, A., Potapenko, A., et al. Highly accurate protein structure prediction with alphafold. nature, 596 0 (7873): 0 583--589, 2021

    2021

  16. [24]

    Elucidating the design space of diffusion-based generative models

    Karras, T., Aittala, M., Aila, T., and Laine, S. Elucidating the design space of diffusion-based generative models. Advances in neural information processing systems, 35: 0 26565--26577, 2022

    2022

  17. [25]

    F., Allain, F

    Klukowski, P., Damberger, F. F., Allain, F. H.-T., Iwai, H., Kadavath, H., Ramelot, T. A., Montelione, G. T., Riek, R., and G \"u ntert, P. The 100-protein nmr spectra dataset: A resource for biomolecular nmr data analysis. Scientific data, 11 0 (1): 0 30, 2024

    2024

  18. [26]

    T., Holton, J

    Lang, P. T., Holton, J. M., Fraser, J. S., and Alber, T. Protein structural ensembles are revealed by redefining x-ray electron density noise. Proceedings of the National Academy of Sciences, 111 0 (1): 0 237--242, 2014

    2014

  19. [27]

    F., Lakomek, N.-A., Fares, C., Schroder, G

    Lange, O. F., Lakomek, N.-A., Fares, C., Schroder, G. F., Walter, K. F., Becker, S., Meiler, J., Grubmuller, H., Griesinger, C., and De Groot, B. L. Recognition dynamics up to microseconds revealed from an rdc-derived ubiquitin ensemble in solution. science, 320 0 (5882): 0 14...

    2008

  20. [28]

    R., Fridovich-Keil, S., Poitevin, F., Zhong, E

    Levy, A., Chan, E. R., Fridovich-Keil, S., Poitevin, F., Zhong, E. D., and Wetzstein, G. Solving inverse problems in protein space using diffusion-based priors. arXiv preprint arXiv:2406.04239, 2024

    2024 arXiv

  21. [29]

    Anisotropic intramolecular backbone dynamics of ubiquitin characterized by nmr relaxation and md computer simulation

    Lienin, S., Bremi, T., Brutscher, B., Br \"u schweiler, R., and Ernst, R. Anisotropic intramolecular backbone dynamics of ubiquitin characterized by nmr relaxation and md computer simulation. Journal of the American Chemical Society, 120 0 (38): 0 9870--9879, 1998

    1998

  22. [30]

    B., DePristo, M

    Lindorff-Larsen, K., Best, R. B., DePristo, M. A., Dobson, C. M., and Vendruscolo, M. Simultaneous determination of protein structure and dynamics. Nature, 433 0 (7022): 0 128--132, 2005

    2005

  23. [31]

    J., Lin, G., Chen, M., Sahoo, A., and Hanson, S

    Liu, Y., Yu, Z., Lindsay, R. J., Lin, G., Chen, M., Sahoo, A., and Hanson, S. M. Exendiff: An experiment-guided diffusion model for protein conformational ensemble generation. bioRxiv, pp.\ 2024--10, 2024

    2024

  24. [32]

    Str2str: A score-based framework for zero-shot protein conformation sampling

    Lu, J., Zhong, B., Zhang, Z., and Tang, J. Str2str: A score-based framework for zero-shot protein conformation sampling. International Conference on Learning Representations, 2014

    2014

  25. [33]

    A., Sellam, N

    Maddipatla, S. A., Sellam, N. B., Vedula, S., Marx, A., and Bronstein, A. Generative modeling of protein ensembles guided by crystallographic electron densities, 2024. URL https://arxiv.org/abs/2412.13223

    2024 arXiv

  26. [34]

    Mallat, S. G. and Zhang, Z. Matching pursuits with time-frequency dictionaries. IEEE Transactions on signal processing, 41 0 (12): 0 3397--3415, 1993

    1993

  27. [35]

    F., Jones, T., Plate, L., Meiler, J., and Gulsevin, A

    McDonald, E. F., Jones, T., Plate, L., Meiler, J., and Gulsevin, A. Benchmarking alphafold2 on peptide structure prediction. Structure, 31 0 (1): 0 111--119, 2023

    2023

  28. [36]

    Colabfold: making protein folding accessible to all

    Mirdita, M., Sch \"u tze, K., Moriwaki, Y., Heo, L., Ovchinnikov, S., and Steinegger, M. Colabfold: making protein folding accessible to all. Nature methods, 19 0 (6): 0 679--682, 2022

    2022

  29. [37]

    Palmer, A. G. NMR characterization of the dynamics of biomacromolecules . Chem. Rev., 104 0 (8): 0 3623--3640, 2004. doi:10.1021/cr030413t

    2004 doi

  30. [38]

    Pytorch: An imperative style, high-performance deep learning library

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 32, 2019

    2019

  31. [39]

    F., Goddard, T

    Pettersen, E. F., Goddard, T. D., Huang, C. C., Meng, E. C., Couch, G. S., Croll, T. I., Morris, J. H., and Ferrin, T. E. Ucsf chimerax: Structure visualization for researchers, educators, and developers. Protein science, 30 0 (1): 0 70--82, 2021

    2021

  32. [40]

    International Tables for Crystallography, Volume C: Mathematical, physical and chemical tables

    Prince, E. International Tables for Crystallography, Volume C: Mathematical, physical and chemical tables. Springer Science & Business Media, 2004

    2004

  33. [41]

    T., Wankowicz, S

    Riley, B. T., Wankowicz, S. A., de Oliveira, S. H., van Zundert, G. C., Hogan, D. W., Fraser, J. S., Keedy, D. A., and van den Bedem, H. qfit 3: Protein and ligand multiconformer modeling for x-ray crystallographic and single-particle cryo-em density maps. Protein Science, 30 ...

    2021

  34. [42]

    A., Marx, A., and Bronstein, A

    Rosenberg, A. A., Marx, A., and Bronstein, A. M. A dataset of alternately located segments in protein crystal structures. Scientific Data, 11 0 (1): 0 783, Jul 2024 a . ISSN 2052-4463. doi:10.1038/s41597-024-03595-4. URL https://doi.org/10.1038/s41597-024-03595-4

    2024 doi

  35. [43]

    A., Vedula, S., Bronstein, A

    Rosenberg, A. A., Vedula, S., Bronstein, A. M., and Marx, A. Seeing double: Molecular dynamics simulations reveal the stability of certain alternate protein conformations in crystal structures. bioRxiv, 2024 b . doi:10.1101/2024.08.31.610605. URL https://www.biorxiv.org/conten...

    2024 doi

  36. [44]

    D., Kuszewski, J

    Schwieters, C. D., Kuszewski, J. J., and Clore, G. M. Using xplor--nih for nmr molecular structure determination. Progress in nuclear magnetic resonance spectroscopy, 48 0 (1): 0 47--62, 2006

    2006

  37. [45]

    Shapovalov, M. V. and Dunbrack, R. L. A smoothed backbone-dependent rotamer library for proteins derived from adaptive kernel density estimates and regressions. Structure, 19 0 (6): 0 844--858, 2011

    2011

  38. [46]

    L., Hendrickson, W

    Smith, J. L., Hendrickson, W. A., Honzatko, R. B., and Sheriff, S. Structural heterogeneity in protein crystals. Biochemistry, 25 0 (18): 0 5018--5027, 1986

    1986

  39. [47]

    and Martin, J

    Smyth, M. and Martin, J. x ray crystallography. Molecular Pathology, 53 0 (1): 0 8, 2000

    2000

  40. [48]

    and Ermon, S

    Song, Y. and Ermon, S. Generative modeling by estimating gradients of the data distribution. Advances in neural information processing systems, 32, 2019

    2019

  41. [49]

    Terrell, G. R. and Scott, D. W. Variable kernel density estimation. The Annals of Statistics, pp.\ 1236--1265, 1992

    1992

  42. [50]

    A., Berger, B., Sander, C., and Marks, D

    Toth-Petroczy, A., Palmedo, P., Ingraham, J., Hopf, T. A., Berger, B., Sander, C., and Marks, D. S. Structured states of disordered proteins from genomic sequences. Cell, 167 0 (1): 0 158--170, 2016

    2016

  43. [51]

    and Fraser, J

    van den Bedem, H. and Fraser, J. S. Integrative, dynamic structural biology at atomic resolution—it's about time . Nat. Meth., 12 0 (4): 0 307--318, 2015. ISSN 1548-7091. doi:10.1038/nmeth.3324. URL http://www.nature.com/nsmb/journal/v13/n3/full/nsmb0306-184.html papers2://pub...

    2015 doi

  44. [52]

    E., and Berendsen, H

    Van Der Spoel, D., Lindahl, E., Hess, B., Groenhof, G., Mark, A. E., and Berendsen, H. J. Gromacs: fast, flexible, and free. Journal of computational chemistry, 26 0 (16): 0 1701--1718, 2005

    2005

  45. [53]

    C., Hudson, B

    van Zundert, G. C., Hudson, B. M., de Oliveira, S. H., Keedy, D. A., Fonseca, R., Heliou, A., Suresh, P., Borrelli, K., Day, T., Fraser, J. S., et al. qfit-ligand reveals widespread conformational heterogeneity of drug-like molecules in x-ray electron density maps. Journal of ...

    2018

  46. [54]

    Attention is all you need

    Vaswani, A. Attention is all you need. Advances in Neural Information Processing Systems, 2017

    2017

  47. [55]

    The nuclear overhauser effect from a quantitative perspective

    V \"o geli, B. The nuclear overhauser effect from a quantitative perspective. Progress in nuclear magnetic resonance spectroscopy, 78: 0 1--46, 2014

    2014

  48. [56]

    M., Caldwell, J

    Wang, J., Wolf, R. M., Caldwell, J. W., Kollman, P. A., and Case, D. A. Development and testing of a general amber force field. Journal of computational chemistry, 25 0 (9): 0 1157--1174, 2004

    2004

  49. [57]

    A., Ravikumar, A., Sharma, S., Riley, B., Raju, A., Hogan, D

    Wankowicz, S. A., Ravikumar, A., Sharma, S., Riley, B., Raju, A., Hogan, D. W., Flowers, J., van den Bedem, H., Keedy, D. A., and Fraser, J. S. Automated multiconformer model building for x-ray crystallography and cryo-em. eLife, 12: 0 RP90606, 2024

    2024

  50. [58]

    M., and Gershoni-Poranne, R

    Weiss, T., Mayo Yanes, E., Chakraborty, S., Cosmo, L., Bronstein, A. M., and Gershoni-Poranne, R. Guided diffusion for inverse molecular design. Nature Computational Science, 3 0 (10): 0 873--882, Oct 2023. ISSN 2662-8457. doi:10.1038/s43588-023-00532-0. URL https://doi.org/10...

    2023 doi

  51. [59]

    Gemmi: A library for structural biology

    Wojdyr, M. Gemmi: A library for structural biology. Journal of Open Source Software, 7 0 (73): 0 4200, 2022

    2022

  52. [60]

    The potts model

    Wu, F.-Y. The potts model. Reviews of modern physics, 54 0 (1): 0 235, 1982

    1982

  53. [61]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.