Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Generative Modeling of Full-Atom Protein Conformations using Latent Diffusion on Graph Embeddings

T0 review · 3 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 LD-FPG generates full-atom conformational ensembles of a membrane protein by learning a compact latent space over Chebyshev graph embeddings of MD snapshots, and the generated ensembles match the reference simulation's all-atom and…

desk verdict A solid, well-documented all-atom ensemble generator for one GPCR, but the headline fidelity numbers are measured in-sample and should be read as trajectory reproduction, not generalization. read the letter →

arxiv 2506.17064 v4 pith:EPJLM7HN submitted 2025-06-20 q-bio.BM cs.LG

Aditya Sengar , Ali Hariri , Daniel Probst , Patrick Barth , Pierre Vandergheynst This is my paper · ORCID
classification q-bio.BMcs.LG
keywords latentdiffusionproteinconformationalensemblesall-atomgenerationChebNetgraphembeddingsGPCRdynamicsdopamineD2receptormolecular
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

LD-FPG asks whether a generative model can replace running new molecular dynamics simulations by learning to sample a specific protein's conformational ensemble at full atomic detail. It encodes rigid-body-aligned MD snapshots of the human dopamine D2 receptor with a Chebyshev graph neural network, compresses the per-atom latent vectors through one of three pooling strategies, trains a diffusion model on the pooled latent space, and decodes new samples back to all-atom Cartesian coordinates. Trained on 12,241 frames from a 2-microsecond simulation in a POPC membrane, the sequential and residue-based variants generate ensembles that match the reference MD distribution closely: all-atom lDDT (a local-distance preservation score) around 0.7, backbone lDDT around 0.8, and individual backbone and side-chain dihedral Jensen–Shannon divergences below 0.03. If this holds, researchers could obtain system-specific, all-atom ensembles of large dynamic targets such as GPCRs without the computational cost of new simulations.

What carries the argument

The central object is the pooled latent deformation space. For each rigid-body-aligned snapshot, an encoder maps $N$ heavy atoms to per-atom vectors $Z(t)\in\mathbb{R}^{N\times d_z}$ using four Chebyshev spectral convolution layers of order $K=4$ on a $k$-nearest-neighbour graph ($k=4$). Three pooling strategies compress these vectors into a low-dimensional diffusion target $h_0$: blind pooling averages globally across all atoms; sequential pooling constructs separate backbone and side-chain contexts; residue-based pooling gives each residue its own local context vector. A DDPM is trained to denoise $h_0$, and a decoder conditioned on the frozen reference latent $Z_{\mathrm{ref}}$ maps sampled latents to coordinates $X_{\mathrm{pred}}$. The load-bearing idea is that conformational diversity can be represented as low-dimensional deformations around a fixed reference fold, so the diffusion model works on a compact latent instead of raw all-atom space.

What would settle it

Retrain the released pipeline on a different D2R MD replica, or on the same replica simulated with a different force field, and compare the generated ensemble against that replica's held-out frames on all-atom lDDT and summed dihedral JSD. If the values fall well below the reported all-atom lDDT of about 0.7 and dihedral JSD below 0.03, the model has memorized the single selected trajectory rather than learned receptor-scale dynamics; if the values hold across replicas, the assumption of representativeness is supported.

Watch

Extended reading notes

Core claim

The central claim is that a complete all-atom conformational ensemble of a membrane protein can be generated from learned latent deformations of a reference structure rather than by predicting coordinates from scratch. The authors train a ChebNet encoder on rigid-body-aligned heavy-atom coordinates from MD snapshots to obtain per-atom latent embeddings, pool these embeddings into a compact latent vector, and train a DDPM to sample that vector; a decoder conditioned on the reference structure's latent embedding maps samples back to coordinates. They report that, for sequential and residue-based pooling, the generated ensemble is statistically close to the MD reference: all-atom lDDT around 0.7, backbone lDDT around 0.8, TM-score above 0.94, and individual backbone and side-chain dihedral JSD values below 0.03. They further report that residue-based pooling gives the best side-chain rotamer distributions, the fewest steric clashes, and the widest coverage of the A100 activation-index landscape, while sequential pooling gives the sharpest backbone geometry. The authors interpret these results as evidence of system-specific ensemble reproduction, including side-chain dynamics, and position LD-FPG as a practical route to all-atom ensemble generation for large, dynamic proteins.

Load-bearing premise

The load-bearing premise is that the single 2-microsecond MD replica chosen as ground truth is representative of the D2 receptor's true conformational landscape; if that trajectory reflects one force field, one starting structure, or one basin, then the generated ensemble faithfully reproduces that trajectory but not the receptor.

Editorial extensions

If this is right

  • Within the studied system, LD-FPG can substitute for additional MD sampling: after training on the 12,241-frame trajectory, new conformations are drawn at far lower cost than microsecond-scale simulation, with ensemble-level statistics close to the reference.
  • Residue-based pooling is the configuration to choose when side-chain rotamers and the A100 activation landscape matter most; it produced the lowest side-chain dihedral JSD, the tightest packing, and the broadest A100 coverage among the three strategies.
  • Sequential pooling is the configuration to choose when backbone geometry matters most; it produced the sharpest Ramachandran distributions and the lowest backbone dihedral JSD after full diffusion generation.
  • The persistent excess of steric clashes over the MD reference (about 1,146 to 1,350 per structure versus about 1,023) is a bottleneck the authors identify, meaning generated structures still need physical-plausibility refinement before direct use in docking or design.
  • The design's conditioning on a reference structure keeps all generated conformations anchored in the same chemical frame, which simplifies learning and makes the framework a plausible template for other large, dynamic proteins with suitable MD data.

Reading between the lines

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

  • The evaluation enforces a distinction between reproducing a single MD trajectory and sampling the receptor's equilibrium ensemble; a natural validation the paper does not report is training the identical pipeline on a second independent D2R replica and checking whether the generated ensemble's distance to the first replica matches the inter-replica distance.
  • Because the starting structure is a ligand-bound cryo-EM state with the ligand removed, the D2R-MD trajectory may be biased toward an apo-like inactive basin; a testable consequence is that retraining on an active-state or ligand-bound trajectory would shift the generated A100 distribution, and the reference-conditioned architecture makes that a direct experiment.
  • Residue pooling's effective latent dimension is about 1,100 per-residue contexts, far larger than the global pooled vectors, which suggests the denoiser capacity, not the latent representation, is the current limit on landscape coverage; a flow-matching or attention-based denoiser over the same pooled representation could plausibly close the gap between decoder and diffusion quality.
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

3 major / 4 minor

Summary. The paper introduces LD-FPG, a latent diffusion framework that generates all-atom heavy-atom conformations of a target protein from MD trajectories. MD snapshots are encoded with a ChebNet into per-atom latents, pooled through one of three strategies (blind, sequential, residue-based), and a DDPM is trained on the pooled latent space; the decoder reconstructs Cartesian coordinates conditioned on the reference frame's latent embedding. On the D2R-MD trajectory of the human dopamine D2 receptor, the authors report all-atom lDDT around 0.7, backbone lDDT around 0.8, and dihedral Jensen-Shannon divergences below 0.03, concluding that LD-FPG reproduces a system-specific MD ensemble without running new simulations.

Significance. If the central claim were validated out-of-sample, LD-FPG would be a practically useful system-specific all-atom ensemble generator for membrane proteins, complementing backbone-only or static predictors. The paper's strengths include public release of code and data, extensive hyperparameter scans in Appendix I, transparent accounting of computational cost, and a careful three-stage decomposition of encoder, decoder, and diffusion errors. However, the significance is currently moderated by the evaluation protocol: all headline metrics are computed against the same trajectory used for training, so the reported numbers measure reproduction of the training distribution rather than generalization to unseen conformations. The method is plausible and the reported numbers are internally consistent, but the evaluation needs substantial strengthening before the main claim can be accepted.

major comments (3)
  1. [Appendix F.1, F.2.2, and I.10] The evaluation protocol is in-sample with respect to the training trajectory. The 12,241 frames are sampled every 100 ps from a single replica, and the chronological 90/10 split places the test frames within the same autocorrelated trajectory as the training frames; for a membrane protein, side-chain and backbone relaxation times are typically tens to hundreds of nanoseconds, so adjacent frames are not independent. The reported lDDT is computed against Xref, which is the first frame and hence a training frame, and the PJSD is computed against the dihedral histogram of the same MD data used for training. These metrics therefore cannot distinguish memorization from generalization. The standard deviations reported in Appendix I.10 are computed over the same temporally correlated frames and do not correct for dependence. The authors should provide autocorrelation times or effective sample sizes, use a decorrelated split, or validate on a second replica or an experimental ensemble.
  2. [Appendix H and Figure 4c] The residue-pooling landscape coverage in Figure 4c is obtained by aggregating samples from 10 distinct diffusion-model checkpoints, as stated in Appendix H. Comparing this aggregated ensemble to the MD ensemble can artificially inflate apparent diversity and A100 coverage, because different checkpoints may specialize to different regions of the latent space. The paper should report per-checkpoint A100 and PCA coverage and justify why multi-epoch aggregation is appropriate for evaluating the distribution produced by a single trained model.
  3. [Appendix F.1] The single-replica MD data was selected post hoc as 'representative' from ten replicas, but no quantitative criterion for representativeness is given. If the chosen replica is not representative of the D2R conformational landscape, then the evaluation target itself is unrepresentative. At minimum, the authors should validate on a second replica or report the selection criterion and the replica-to-replica variance of the key structural metrics.
minor comments (4)
  1. [Abstract] The abstract says a 2 microsecond MD trajectory with about 12,000 frames, while Appendix F.1 reports 12,241 frames from a 1.224 microsecond analysis trajectory after discarding 776 ns; please reconcile these numbers.
  2. [Tables 1 and 2] The main tables report mean values only; since Appendix I.10 provides standard deviations, the main text should either include them or explicitly point to Appendix I.10 so that readers are not misled into overprecision.
  3. [Section 4.2 and Figure 4] The main text should state that the residue-pooling panels in Figure 4 aggregate samples from 10 checkpoints; the current wording in Section 4.2 mentions multi-epoch sampling only parenthetically, with the full explanation confined to Appendix H.
  4. [Title and Section 3.1] The term 'full-atom' may be misleading because the model operates on heavy atoms only and explicitly excludes hydrogen atoms; consider qualifying the title or the relevant methodology text as 'all heavy-atom' to avoid overstatement.

Circularity Check

0 steps flagged · score 0.0 of 10

No formal circularity: the method is an empirical fit to MD data and is evaluated on a held-out chronological split, not on a derived quantity.

full rationale

The paper's derivation chain is self-contained in the sense required here. LD-FPG is an empirical generative model trained on MD frames; no analytic prediction is derived from a fitted parameter, and no equation in the paper reduces to its own input by construction. The ChebNet encoder, pooling decoders, and DDPM are trained with MSE objectives on a training split, and the reported lDDT/PJSD numbers are computed on a held-out chronological test split (Appendix F.1), so the headline ensemble-fidelity claim is a model-fit assessment rather than a derivation from an input. The only quasi-circular feature is that the reference ensemble used both for training and for evaluation is the same D2R-MD trajectory; this is an in-sample/autocorrelation validity concern, not a formal circularity, because the test frames are not the training frames and the evaluation metrics are not the training losses. No self-citation is load-bearing: ChebNet is cited as an external architecture, and the Barth-group background citations are contextual. There is no invoked uniqueness theorem, no ansatz smuggled in via citation, and no renaming of a known result. Therefore no circular step can be quoted and no circularity score above 0 is justified.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

No new physical entities are introduced; the latent vectors are learned internal representations, not postulates about nature. The load-bearing assumptions are the representativeness of the MD data, the sufficiency of the graph encoding and pooling, and the adequacy of the evaluation metrics.

free parameters (5)
  • Encoder latent dimension dz per pooling strategy = blind 16, sequential 8, residue 4
    Tuned per pooling strategy; larger dz improves atom-wise reconstruction but makes pooled latent diffusion harder. Reported in Sections 3.2 and I.5.
  • Pooled latent dimension dp = blind 100; sequential backbone 135 and sidechain 108; residue 4 per residue (1092 total)
    Chosen from scans; dp above 200-300 hindered diffusion training and dp below 50 degraded reconstruction (Section 3.3).
  • Diffusion schedule and denoiser size = T=500 (blind/sequential) to 1500 (residue); beta_start 5e-6 or 0.005; beta_end 0.02 or 0.1; hidden dims 1024 or 4096
    Grid-searched in Appendix I.6 and I.8; these choices affect sampling quality and diversity.
  • Dihedral fine-tuning weights for blind pooling = wbase=1.0, lambda_div=2.0, lambda_mse=0.0, fdih=0.1
    Chosen for optional fine-tuning; dihedral losses improve JSD marginally but slightly reduce lDDT (Tables 1 and 2).
  • Graph construction and ChebNet order = k=4 nearest neighbors, Chebyshev order K=4, 4 layers
    Chosen as a performance/compute balance; ablations in I.4 show other settings give similar or better reconstruction loss, so this choice is not uniquely determined.
assumptions (6)
  • domain assumption The selected D2R-MD trajectory is a representative sample of the D2R equilibrium conformational ensemble.
    The entire training and evaluation pipeline treats this single 1.224 microsecond trajectory, from one of ten replicas, as ground truth. No cross-replica or experimental validation is provided. (Appendix F.1)
  • standard math Rigid-body Kabsch alignment to the first frame removes only global rotation and translation, leaving internal conformational signal intact.
    Standard assumption in trajectory analysis; used in all preprocessing and metric computation. (Section 3.1)
  • domain assumption A k-nearest-neighbor graph with k=4 and ChebNet order K=4 captures local atomic geometry sufficiently for reconstruction.
    Supported by encoder ablations, but sufficiency is only demonstrated on this D2R system. (Appendix I.4)
  • domain assumption The pooled latent representation h0 retains enough information for the decoder to reconstruct all-atom coordinates.
    Reconstruction from ground-truth latents is the upper bound; pooling is a lossy compression, and the trade-off is tuned for D2R only. (Section 3.3)
  • domain assumption The DDPM trained on finite pooled latents learns the true underlying distribution, and ancestral sampling yields novel representative latents.
    Standard DDPM assumption; its validity here is assessed only via distributional distance to the same MD trajectory. (Section 3.4)
  • domain assumption lDDT, TM-score, dihedral JSD, clash count, and A100 collectively measure ensemble fidelity and physical plausibility.
    These are accepted protein-structure metrics, but no single metric validates the functional relevance of the generated ensembles. (Appendix F.2)

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generative Modeling of Full-Atom Protein Conformations using Latent Diffusion on Graph Embeddings." pith.science (2026). https://pith.science/paper/EPJLM7HN

@misc{pith2026250617064,
  author       = {Pith},
  title        = {Pith review of: Generative Modeling of Full-Atom Protein Conformations using Latent Diffusion on Graph Embeddings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EPJLM7HN}},
  note         = {Machine review of arXiv:2506.17064}
}
read the original abstract

Generating diverse, all-atom conformational ensembles of dynamic proteins such as G-protein-coupled receptors (GPCRs) is critical for understanding their function, yet most generative models simplify atomic detail or ignore conformational diversity altogether. We present latent diffusion for full protein generation (LD-FPG), a framework that constructs complete all-atom protein structures, including every side-chain heavy atom, directly from molecular dynamics (MD) trajectories. LD-FPG employs a Chebyshev graph neural network (ChebNet) to obtain low-dimensional latent embeddings of protein conformations, which are processed using three pooling strategies: blind, sequential and residue-based. A diffusion model trained on these latent representations generates new samples that a decoder, optionally regularized by dihedral-angle losses, maps back to Cartesian coordinates. Using D2R-MD, a 2-microsecond MD trajectory (12 000 frames) of the human dopamine D2 receptor in a membrane environment, the sequential and residue-based pooling strategy reproduces the reference ensemble with high structural fidelity (all-atom lDDT of approximately 0.7; C-alpha-lDDT of approximately 0.8) and recovers backbone and side-chain dihedral-angle distributions with a Jensen-Shannon divergence of less than 0.03 compared to the MD data. LD-FPG thereby offers a practical route to system-specific, all-atom ensemble generation for large proteins, providing a promising tool for structure-based therapeutic design on complex, dynamic targets. The D2R-MD dataset and our implementation are freely available to facilitate further research.

Figures

Figures reproduced from arXiv: 2506.17064 by the authors.

Figure 1
Figure 1. Schematic of the LD-FPG framework. The workflow depicts: (1) ChebNet encoding of MD frames to atom-wise latent embeddings (Z (t) ). (2) Pooling of Z (t) into a compact latent context (h0). (3) DDPM-based sampling of this pooled latent space to generate h gen 0 . (4) Decoding of h gen 0 , conditioned on reference latents (Zref), to all-atom coordinates (Xpred). The conceptual details of the framework are described in… view at source ↗
Figure 2
Figure 2. Comparison of dihedral angle distributions for different pooling strategies. (Top row) [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Examples of generated D2R conformations using different pooling strategies. Panels [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Latent space and collective variable (A100) analysis for Blind (a), Sequential (b), and [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Spectral Diffusion for Protein Dynamics

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

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

Reference graph

Works this paper leans on

107 extracted references · 62 canonical work pages · cited by 1 Pith paper

  1. [1]

    Molecular dynamics and protein function

    Martin Karplus and John Kuriyan. Molecular dynamics and protein function. Proceedings of the National Academy of Sciences, 102(19):6679–6685, 2005

    2005

  2. [2]

    Dynamic personalities of proteins

    Katherine Henzler-Wildman and Dorothee Kern. Dynamic personalities of proteins. Nature, 450(7172):964–972, 2007

    2007

  3. [3]

    Highly accurate protein structure prediction with AlphaFold

    John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Olaf Ron- neberger, Kathryn Tunyasuvunakool, Russ Bates, Augustin ˇZ´ıdek, Anna Potapenko, et al. Highly accurate protein structure prediction with AlphaFold. Nature, 596(7873):583–589, 2021

    2021

  4. [4]

    Accurate pre- diction of protein structures and interactions using a three-track neural network

    Minkyung Baek, Frank DiMaio, Ivan Anishchenko, Justas Dauparas, Sergey Ovchinnikov, Gyu Rie Lee, Jue Wang, Qian Cong, Lisa N Kinch, R Dustin Schaeffer, et al. Accurate pre- diction of protein structures and interactions using a three-track neural network. Science, 373 (6557):871–876, 2021

    2021

  5. [5]

    Evolutionary-scale prediction of atomic-level protein structure with a language model

    Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Nikita Smetanin, Robert Verkuil, Ori Kabeli, Yaniv Shmueli, et al. Evolutionary-scale prediction of atomic-level protein structure with a language model. Science, 379(6637):1123–1130, 2023

    2023

  6. [6]

    Accurate structure prediction of biomolecular interactions with alphafold 3

    Josh Abramson, Jonas Adler, Jack Dunger, Richard Evans, Tim Green, Alexander Pritzel, Olaf Ronneberger, Lindsay Willmore, Andrew J Ballard, Joshua Bambrick, et al. Accurate structure prediction of biomolecular interactions with alphafold 3. Nature, 630(8016):493–500, 2024

    2024

  7. [7]

    Boltz-1: Democratizing biomolecular interaction modeling

    Jeremy Wohlwend, Gabriele Corso, Saro Passaro, Mateo Reveiz, Ken Leidal, Wojtek Swider- ski, Tally Portnoi, Itamar Chinn, Jacob Silterra, Tommi Jaakkola, et al. Boltz-1: Democratizing biomolecular interaction modeling. bioRxiv, pages 2024–11, 2024. doi: 10.1101/2024.11.08 .622384

    work page doi:10.1101/2024.11.08 2024

  8. [8]

    The role of dynamic conformational ensembles in biomolecular recognition

    David D Boehr, Ruth Nussinov, and Peter E Wright. The role of dynamic conformational ensembles in biomolecular recognition. Nature chemical biology, 5(11):789–796, 2009

    2009

Show all 107 references
  1. [9]

    Implications of protein flexibility for drug discovery

    Simon J Teague. Implications of protein flexibility for drug discovery. Nature reviews Drug discovery, 2(7):527–541, 2003

    2003

  2. [10]

    Computational design of g protein- coupled receptor allosteric signal transductions

    Kuang-Yui Michael Chen, Daniel Keri, and Patrick Barth. Computational design of g protein- coupled receptor allosteric signal transductions. Nature Chemical Biology, 16(1):77–86, 2020

    2020

  3. [11]

    Computational design of highly signalling-active membrane receptors through solvent-mediated allosteric networks

    K-YM Chen, JK Lai, LSP Rudden, J Wang, AM Russell, K Conners, ME Rutter, B Condon, F Tung, L Kodandapani, et al. Computational design of highly signalling-active membrane receptors through solvent-mediated allosteric networks. Nature Chemistry, pages 1–10, 2025

    2025

  4. [12]

    Side-chain flex- ibility in proteins upon ligand binding

    Rafael Najmanovich, Josef Kuttner, Vladimir Sobolev, and Marvin Edelman. Side-chain flex- ibility in proteins upon ligand binding. Proteins: Structure, Function, and Bioinformatics, 39 (3):261–268, 2000

    2000

  5. [13]

    Deep learning approaches for confor- mational flexibility and switching properties in protein design

    Lucas SP Rudden, Mahdi Hijazi, and Patrick Barth. Deep learning approaches for confor- mational flexibility and switching properties in protein design. Frontiers in Molecular Bio- sciences, 9:928534, 2022

    2022

  6. [14]

    De novo design of protein structure and function with rfdiffusion.Nature, 620(7976):1089–1100, 2023

    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, et al. De novo design of protein structure and function with rfdiffusion.Nature, 620(7976):1089–1100, 2023

    2023

  7. [15]

    Protein structure generation via folding diffusion

    Kevin E Wu, Kevin K Yang, Rianne van den Berg, Sarah Alamdari, James Y Zou, Alex X Lu, and Ava P Amini. Protein structure generation via folding diffusion. Nature communications, 15(1):1059, 2024

    2024

  8. [16]

    Pro- teina: Scaling flow-based protein structure generative models

    Tomas Geffner, Kieran Didi, Zuobai Zhang, Danny Reidenbach, Zhonglin Cao, Jason Yim, Mario Geiger, Christian Dallago, Emine Kucukbenli, Arash Vahdat, and Karsten Kreis. Pro- teina: Scaling flow-based protein structure generative models. In International Conference on Learning ...

    2025

  9. [17]

    An all-atom protein generative model

    Alexander E Chu, Jinho Kim, Lucy Cheng, Gina El Nesr, Minkai Xu, Richard W Shuai, and Po-Ssu Huang. An all-atom protein generative model. Proceedings of the National Academy of Sciences, 121(27):e2311500121, 2024

    2024

  10. [18]

    Illumi- nating protein space with a programmable generative model

    John B Ingraham, Max Baranov, Zak Costello, Karl W Barber, Wujie Wang, Ahmed Ismail, Vincent Frappier, Dana M Lord, Christopher Ng-Thow-Hing, Erik R Van Vlack, et al. Illumi- nating protein space with a programmable generative model. Nature, 623(7989):1070–1078, 2023

    2023

  11. [19]

    Alphafold2-rave: From sequence to boltzmann ranking

    Bodhi P Vani, Akashnathan Aranganathan, Dedi Wang, and Pratyush Tiwary. Alphafold2-rave: From sequence to boltzmann ranking. Journal of chemical theory and computation , 19(14): 4351–4354, 2023

    2023

  12. [20]

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

    Jiarui Lu, Bozitao Zhong, Zuobai Zhang, and Jian Tang. Str2str: A score-based framework for zero-shot protein conformation sampling. arXiv preprint arXiv:2306.03117, 2023

    2023 arXiv

  13. [21]

    Alphafold meets flow matching for gener- ating protein ensembles

    Bowen Jing, Bonnie Berger, and Tommi Jaakkola. Alphafold meets flow matching for gener- ating protein ensembles. arXiv preprint arXiv:2402.04845, 2024

    2024 arXiv

  14. [22]

    Sarah Lewis, Tim Hempel, Jos ´e Jim ´enez-Luna, Michael Gastegger, Yu Xie, Andrew Y . K. Foong, Victor Garcia Satorras, Osama Abdin, Bastiaan S. Veeling, et al. Scalable emulation of protein equilibrium ensembles with generative deep learning. bioRxiv, Dec 2024. doi: 10.1101/2...

    2024 doi

  15. [23]

    Predicting equilibrium distributions for molecular systems with deep learning

    Shuxin Zheng, Jiyan He, Chang Liu, Yu Shi, Ziheng Lu, Weitao Feng, Fusong Ju, Jiaxi Wang, Jianwei Zhu, Yaosen Min, et al. Predicting equilibrium distributions for molecular systems with deep learning. Nature Machine Intelligence, 6(5):558–567, 2024

    2024

  16. [24]

    A latent diffusion model for protein structure generation

    Cong Fu, Keqiang Yan, Limei Wang, Wing Yee Au, Michael Curtis McThrow, Tao Komikado, Koji Maruhashi, Kanji Uchino, Xiaoning Qian, and Shuiwang Ji. A latent diffusion model for protein structure generation. In Learning on Graphs Conference, pages 29–1. PMLR, 2024

    2024

  17. [25]

    Lu, Wilson Yan, Vladimir Gligorijevic, Kyunghyun Cho, Richard Bonneau, Kevin K

    Amy X. Lu, Wilson Yan, Vladimir Gligorijevic, Kyunghyun Cho, Richard Bonneau, Kevin K. Yang, Pieter Abbeel, and Nathan C. Frey. Generating all-atom protein structure from sequence- only training data. bioRxiv preprint , 2024. doi: 10.1101/2024.12.19.630448. doi: 10.1101/2024.1...

    2024 doi

  18. [26]

    Transferable deep generative modeling of intrinsically disordered protein conformations

    Giacomo Janson and Michael Feig. Transferable deep generative modeling of intrinsically disordered protein conformations. PLoS Computational Biology, 20(5):e1012144, 2024

    2024

  19. [27]

    Protein conformation generation via force-guided SE(3) diffusion models

    Yan Wang, Lihao Wang, Yuning Shen, Yiqun Wang, Huizhuo Yuan, Yue Wu, and Quanquan Gu. Protein conformation generation via force-guided SE(3) diffusion models. In Proceedings of the 41st International Conference on Machine Learning , volume 235 of Proceedings of Machine Learnin...

    2024 arXiv

  20. [28]

    Yikai Liu, Abhilash Sahoo, Zongxin Yu, Guang Lin, Ming Chen, and Sonya M. Hanson. EGDiff: An experiment-guided diffusion model for protein conformational ensemble gener- ation. bioRxiv preprint, 2024. doi: 10.1101/2024.10.04.616517. version 1 posted Oct 4 2024

    2024 doi

  21. [29]

    Latorraca, A

    Naomi R. Latorraca, A. J. Venkatakrishnan, and Ron O. Dror. GPCR dynamics: Structures in motion. Chemical Reviews, 117(1):139–155, 2017

    2017

  22. [30]

    G protein-coupled receptors (gpcrs): advances in structures, mechanisms and drug discovery

    Mingyang Zhang, Ting Chen, Xun Lu, Xiaobing Lan, Ziqiang Chen, and Shaoyong Lu. G protein-coupled receptors (gpcrs): advances in structures, mechanisms and drug discovery. Signal Transduction and Targeted Therapy, 9(1), 2024. ISSN 2059-3635. doi: 10.1038/s413 92-024-01803-6. U...

    2024 doi

  23. [31]

    What are the current trends in g protein-coupled receptor targeted drug discovery? Expert Opinion on Drug Discovery , 18(8):815–820, 2023

    Vicent Casad ´o and Ver`onica Casad´o-Anguera. What are the current trends in g protein-coupled receptor targeted drug discovery? Expert Opinion on Drug Discovery , 18(8):815–820, 2023. ISSN 1746-045X. doi: 10.1080/17460441.2023.2216014. URL http://dx.doi.org/10. 1080/17460441...

    2023

  24. [32]

    G protein-coupled receptors: structure- and function- based drug discovery

    Dehua Yang, Qingtong Zhou, Viktorija Labroska, Shanshan Qin, Sanaz Darbalaei, Yiran Wu, Elita Yuliantie, Linshan Xie, Houchao Tao, Jianjun Cheng, Qing Liu, Suwen Zhao, Wenqing Shui, Yi Jiang, and Ming-Wei Wang. G protein-coupled receptors: structure- and function- based drug d...

    2021

  25. [33]

    Drugbank 5.0: a major update to the drugbank database for 2018

    David S Wishart, Yannick D Feunang, An C Guo, Elvis J Lo, Ana Marcu, Jason R Grant, Tanvir Sajed, Daniel Johnson, Carin Li, Zinat Sayeeda, et al. Drugbank 5.0: a major update to the drugbank database for 2018. Nucleic acids research, 46(D1):D1074–D1082, 2018

    2018

  26. [34]

    Structure and dynamics of gpcr signaling complexes

    Daniel Hilger, Matthieu Masureel, and Brian K Kobilka. Structure and dynamics of gpcr signaling complexes. Nature structural & molecular biology, 25(1):4–12, 2018

    2018

  27. [35]

    Monod, J

    J. Monod, J. Wyman, and J.-P. Changeux. On the nature of allosteric transitions: A plausible model. Journal of Molecular Biology, 12(1):88–118, 1965

    1965

  28. [36]

    Gpcr dynamics: structures in motion

    Naomi R Latorraca, AJ Venkatakrishnan, and Ron O Dror. Gpcr dynamics: structures in motion. Chemical reviews, 117(1):139–155, 2017

    2017

  29. [37]

    Computational design of dynamic receptor—peptide signaling complexes applied to chemotaxis

    Robert E Jefferson, Aur ´elien Oggier, Andreas F ¨uglistaler, Nicolas Camviel, Mahdi Hijazi, Ana Rico Villarreal, Caroline Arber, and Patrick Barth. Computational design of dynamic receptor—peptide signaling complexes applied to chemotaxis. Nature Communications, 14 (1):2875, 2023

    2023

  30. [38]

    Biased receptor signaling in drug discovery

    Terry Kenakin. Biased receptor signaling in drug discovery. Pharmacological reviews, 71(2): 267–315, 2019

    2019

  31. [39]

    Goupil, S

    E. Goupil, S. A. Laporte, and T. E. Hebert. Functional selectivity in gpcr signaling: Un- derstanding the full spectrum of receptor conformations. Mini-Reviews in Medicinal Chem- istry, 12(9):817–830, 2012. ISSN 1389-5575. doi: 10.2174/138955712800959143. URL http://dx.doi.org...

    2012 doi

  32. [40]

    Jeffrey Conn, Arthur Christopoulos, and Craig W

    P. Jeffrey Conn, Arthur Christopoulos, and Craig W. Lindsley. Allosteric modulators of gpcrs: a novel approach for the treatment of cns disorders. Nature Reviews Drug Discovery , 8(1): 41–54, 2009. ISSN 1474-1784. doi: 10.1038/nrd2760. URL http://dx.doi.org/10.1038 /nrd2760

    2009 doi

  33. [41]

    Deep learning dynamic allostery of G-Protein- Coupled receptors

    Hung N Do, Jinan Wang, and Yinglong Miao. Deep learning dynamic allostery of G-Protein- Coupled receptors. JACS Au, 3(11):3165–3180, 2023

    2023

  34. [42]

    Can molecular dynamics simulations improve the structural accuracy and virtual screening perfor- mance of gpcr models? PLOS Computational Biology, 17(5):e1008936, 2021

    Jon Kapla, Ismael Rodr ´ıguez-Espigares, Flavio Ballante, Jana Selent, and Jens Carlsson. Can molecular dynamics simulations improve the structural accuracy and virtual screening perfor- mance of gpcr models? PLOS Computational Biology, 17(5):e1008936, 2021. ISSN 1553-

    2021

  35. [43]

    Eric Xu, and Xi Cheng

    Xin-heng He, Chong-zhao You, Hua-liang Jiang, Yi Jiang, H. Eric Xu, and Xi Cheng. Al- phafold2 versus experimental structures: evaluation on g protein-coupled receptors.Acta Phar- macologica Sinica, 44(1):1–7, 2022. ISSN 1745-7254. doi: 10.1038/s41401-022-00938-y. URL http://d...

    2022 doi

  36. [44]

    Gpcrmd uncovers the dynamics of the 3d-gpcrome

    Ismael Rodr ´ıguez-Espigares, Mariona Torrens-Fontanals, Johanna KS Tiemann, David Aranda-Garc´ıa, Juan Manuel Ram ´ırez-Anguita, Tomasz Maciej Stepniewski, Nathalie Worp, Alejandro Varela-Rial, Adri´an Morales-Pastor, Brian Medel-Lacruz, et al. Gpcrmd uncovers the dynamics of...

    2020

  37. [45]

    Gpcr molecular dynamics forecasting using recur- rent neural networks

    J M L ´opez-Correa, C K¨onig, and A Vellido. Gpcr molecular dynamics forecasting using recur- rent neural networks. Scientific Reports, 14(1):9942, 2024. doi: 10.1038/s41598-023-48346-4

    2024 doi

  38. [46]

    Convolutional neural networks on graphs with fast localized spectral filtering

    Micha ¨el Defferrard, Xavier Bresson, and Pierre Vandergheynst. Convolutional neural networks on graphs with fast localized spectral filtering. Advances in neural information processing systems, 29, 2016. 13

    2016

  39. [47]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems, volume 33, pages 6840–6851, 2020

    2020

  40. [48]

    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

  41. [49]

    Sequence-augmented se (3)-flow matching for conditional protein backbone generation

    Guillaume Huguet, James Vuckovic, Kilian Fatras, Eric Thibodeau-Laufer, Pablo Lemos, Ri- ashat Islam, Cheng-Hao Liu, Jarrid Rector-Brooks, Tara Akhound-Sadegh, Michael Bronstein, et al. Sequence-augmented se (3)-flow matching for conditional protein backbone generation. arXiv ...

    2024 arXiv

  42. [50]

    Ig-vae: Generative modeling of protein structure by direct 3d coordinate generation

    Raphael R Eguchi, Christian A Choe, and Po-Ssu Huang. Ig-vae: Generative modeling of protein structure by direct 3d coordinate generation. PLoS computational biology , 18(6): e1010271, 2022

    2022

  43. [51]

    FlowPacker: Protein side-chain packing with torsional flow matching

    Jin Sub Lee and Philip M Kim. FlowPacker: Protein side-chain packing with torsional flow matching. Bioinformatics, page btaf010, 2025. doi: 10.1093/bioinformatics/btaf010

    2025 doi

  44. [52]

    Predicting mutational effects on protein-protein binding via a side-chain diffusion probabilistic model

    Shiwei Liu, Tian Zhu, Milong Ren, Chungong Yu, Dongbo Bu, and Haicang Zhang. Predicting mutational effects on protein-protein binding via a side-chain diffusion probabilistic model. Advances in Neural Information Processing Systems, 36:48994–49005, 2023

    2023

  45. [53]

    Protein en- semble generation through variational autoencoder latent space sampling.Journal of Chemical Theory and Computation, 20(7):2689–2695, 2024

    Sanaa Mansoor, Minkyung Baek, Hahnbeom Park, Gyu Rie Lee, and David Baker. Protein en- semble generation through variational autoencoder latent space sampling.Journal of Chemical Theory and Computation, 20(7):2689–2695, 2024

    2024

  46. [54]

    P2dflow: A protein ensemble generative model with SE(3) flow matching

    Yaowei Jin, Qi Huang, Ziyang Song, Mingyue Zheng, Dan Teng, and Qian Shi. P2dflow: A protein ensemble generative model with SE(3) flow matching. Journal of Chemical Theory and Computation, 21(6):3288–3296, 2025. doi: 10.1021/acs.jctc.4c01620

    2025 doi

  47. [55]

    Generative modeling of molecular dynamics trajectories

    Bowen Jing, Bonnie Berger, and Tommi Jaakkola. Generative modeling of molecular dynamics trajectories. arXiv preprint arXiv:2409.17808 , 2024. NeurIPS 2024 ML4LMS Workshop Poster

    2024 arXiv

  48. [56]

    A solution for the best rotation to relate two sets of vectors

    Wolfgang Kabsch. A solution for the best rotation to relate two sets of vectors. Acta Crystal- lographica Section A: Crystal Physics, Diffraction, Theoretical and General Crystallography, 32(5):922–923, 1976

    1976

  49. [57]

    Universal activation index for class a gpcrs

    Passainte Ibrahim, David Wifling, and Timothy Clark. Universal activation index for class a gpcrs. Journal of Chemical Information and Modeling, 59(9):3938–3945, 2019

    2019

  50. [58]

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

    A Paszke. Pytorch: An imperative style, high-performance deep learning library. arXiv preprint arXiv:1912.01703, 2019

    1912 arXiv

  51. [59]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  52. [60]

    Diffpie: Guiding deep generative models to explore protein conformations under external interactions

    Yanbin Wang and Ming Chen. Diffpie: Guiding deep generative models to explore protein conformations under external interactions. bioRxiv, pages 2025–04, 2025

    2025

  53. [61]

    Physdock: A physics-guided all-atom diffusion model for protein-ligand complex prediction

    Kexin Zhang, Yuanyuan Ma, Jiale Yu, Huiting Luo, Jinyu Lin, Yifan Qin, Xiangcheng Li, Qian Jiang, Fang Bai, Jiayi Dou, et al. Physdock: A physics-guided all-atom diffusion model for protein-ligand complex prediction. bioRxiv, pages 2025–04, 2025

    2025

  54. [62]

    Petimot: A novel framework for infer- ring protein motions from sparse data using se (3)-equivariant graph neural networks

    Valentin Lombard, Sergei Grudinin, and Elodie Laine. Petimot: A novel framework for infer- ring protein motions from sparse data using se (3)-equivariant graph neural networks. arXiv preprint arXiv:2504.02839, 2025

    2025 arXiv

  55. [63]

    Atomica: Learning universal representations of intermolecular interactions

    Ada Fang, Zaixi Zhang, Andrew Zhou, and Marinka Zitnik. Atomica: Learning universal representations of intermolecular interactions. bioRxiv, pages 2025–04, 2025

    2025

  56. [64]

    All-atom diffusion transformers: Unified generative modelling of molecules and materials

    Chaitanya K Joshi, Xiang Fu, Yi-Lun Liao, Vahe Gharakhanyan, Benjamin Kurt Miller, Anuroop Sriram, and Zachary W Ulissi. All-atom diffusion transformers: Unified generative modelling of molecules and materials. 2025. 14

    2025

  57. [65]

    P2DFlow: A protein ensemble generative model with SE(3) flow matching

    Yangtian Zhan, Zuobai Zhang, Tianfan Li, Yu Tian, Hao Li, Sheng Wang, and Jian Xu. P2DFlow: A protein ensemble generative model with SE(3) flow matching. arXiv preprint arXiv:2411.17196, 2024

    2024 arXiv

  58. [66]

    Structure of the d2 dopamine receptor bound to the atypical antipsychotic drug risperidone

    Sheng Wang, Tao Che, Anat Levit, Brian K Shoichet, Daniel Wacker, and Bryan L Roth. Structure of the d2 dopamine receptor bound to the atypical antipsychotic drug risperidone. Nature, 555(7695):269–273, 2018

    2018

  59. [67]

    Rosettaremodel: a generalized framework for flexible backbone protein design

    Po-Ssu Huang, Yih-En Andrew Ban, Florian Richter, Ingemar Andre, Robert Vernon, William R Schief, and David Baker. Rosettaremodel: a generalized framework for flexible backbone protein design. PloS one, 6(8):e24109, 2011

    2011

  60. [68]

    Charmm-gui membrane builder toward realistic biological membrane simulations, 2014

    Emilia L Wu, Xi Cheng, Sunhwan Jo, Huan Rui, Kevin C Song, Eder M D ´avila-Contreras, Yifei Qi, Jumin Lee, Viviana Monje-Galvan, Richard M Venable, et al. Charmm-gui membrane builder toward realistic biological membrane simulations, 2014

    2014

  61. [69]

    Structure and dynamics of the tip3p, spc, and spc/e water models at 298 k

    Pekka Mark and Lennart Nilsson. Structure and dynamics of the tip3p, spc, and spc/e water models at 298 k. The Journal of Physical Chemistry A, 105(43):9954–9960, 2001

    2001

  62. [70]

    Charmm36m: an improved force field for folded and intrinsically disordered proteins

    Jing Huang, Sarah Rauscher, Grzegorz Nawrocki, Ting Ran, Michael Feig, Bert L De Groot, Helmut Grubm¨uller, and Alexander D MacKerell Jr. Charmm36m: an improved force field for folded and intrinsically disordered proteins. Nature methods, 14(1):71–73, 2017

    2017

  63. [71]

    Gromacs: High performance molecular simulations through multi- level parallelism from laptops to supercomputers

    Mark James Abraham, Teemu Murtola, Roland Schulz, Szil ´ard P ´all, Jeremy C Smith, Berk Hess, and Erik Lindahl. Gromacs: High performance molecular simulations through multi- level parallelism from laptops to supercomputers. SoftwareX, 1:19–25, 2015

    2015

  64. [72]

    Canonical sampling through veloc- ity rescaling

    Giovanni Bussi, Davide Donadio, and Michele Parrinello. Canonical sampling through veloc- ity rescaling. The Journal of chemical physics, 126(1), 2007

    2007

  65. [73]

    Pressure control using stochastic cell rescaling

    Mattia Bernetti and Giovanni Bussi. Pressure control using stochastic cell rescaling. The Journal of Chemical Physics, 153(11), 2020

    2020

  66. [74]

    Lincs: A linear constraint solver for molecular simulations.Journal of computational chemistry, 18(12):1463– 1472, 1997

    Berk Hess, Henk Bekker, Herman JC Berendsen, and Johannes GEM Fraaije. Lincs: A linear constraint solver for molecular simulations.Journal of computational chemistry, 18(12):1463– 1472, 1997

    1997

  67. [75]

    A smooth particle mesh ewald method

    Ulrich Essmann, Lalith Perera, Max L Berkowitz, Tom Darden, Hsing Lee, and Lee G Ped- ersen. A smooth particle mesh ewald method. The Journal of chemical physics , 103(19): 8577–8593, 1995

    1995

  68. [76]

    lddt: a local superposition-free score for comparing protein structures and models using distance difference tests

    Valerio Mariani, Marco Biasini, Alessandro Barbato, and Torsten Schwede. lddt: a local superposition-free score for comparing protein structures and models using distance difference tests. Bioinformatics, 29(21):2722–2728, 2013

    2013

  69. [77]

    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(7):2302–2309, 2005

    2005

  70. [78]

    Biopython: freely available python tools for computational molecular biology and bioinformatics

    Peter JA Cock, Tiago Antao, Jeffrey T Chang, Brad A Chapman, Cymon J Cox, Andrew Dalke, Iddo Friedberg, Thomas Hamelryck, Frank Kauff, Bartek Wilczynski, et al. Biopython: freely available python tools for computational molecular biology and bioinformatics. Bioinformat- ics, 2...

    2009

  71. [79]

    Oliphant, Matt Haberland, Tyler Reddy, David Cour- napeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, St ´efan J

    Pauli Virtanen, Ralf Gommers, Travis E. Oliphant, Matt Haberland, Tyler Reddy, David Cour- napeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, St ´efan J. van der Walt, Matthew Brett, Joshua Wilson, K. Jarrod Millman, Nikolay Mayorov, Andrew R. J. Nelso...

    2020

  72. [80]

    Boosting diffusion models with moving average sampling in frequency domain

    Yurui Qian, Qi Cai, Yingwei Pan, Yehao Li, Ting Yao, Qibin Sun, and Tao Mei. Boosting diffusion models with moving average sampling in frequency domain. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8911–8920, 2024

    2024

  73. [81]

    Towards the systematic reporting of the energy and carbon footprints of machine learning

    Peter Henderson, Jieru Hu, Joshua Romoff, Emma Brunskill, Dan Jurafsky, and Joelle Pineau. Towards the systematic reporting of the energy and carbon footprints of machine learning. Journal of Machine Learning Research, 21(248):1–43, 2020

    2020

  74. [82]

    https://www.nvidia.com/en-us/data-center/l40s/

  75. [83]

    https://www.hyperstack.cloud/technical-resources/performance-benchmark s/comparing-nvidia-h100-pcie-vs-sxm-performance-use-cases-and-more

  76. [84]

    https://journal.uptimeinstitute.com/large-data-centers-are-mostly-mor e-efficient-analysis-confirms/

  77. [85]

    https://ember-energy.org/chapter/electricity-transition-in-2022-ger2023 /

    2022

  78. [86]

    https://aliunid.com/wp-content/uploads/2023/04/2023-media-information-a liunid-CO2-content-in-Swiss-electricity.pdf . 16 A Methodology Overview and Notation This section provides a consolidated overview of the LD-FPG framework’s workflow, key mathe- matical notation used throu...

    2023

  79. [89]

    This operation pools across all N atoms for each sample in the batch: h(b) global =Pglobal(Z(b))∈ Rdp wheredp =H·W is the dimension of the pooled global context vector

    Global Pooling: The atom embeddings Z(b) are treated as an image-like tensor and pro- cessed by a 2D adaptive average pooling layer,Pglobal (typically nn.AdaptiveAvgPool2d with output sizeH×W ). This operation pools across all N atoms for each sample in the batch: h(b) global ...

  80. [90]

    For a batch, this is Hglobal ex∈ RB×N×dp

    Context Expansion: This global context vector h(b) global is then expanded (tiled) to match the number of atoms N, resulting in H(b) global ex∈ RN×dp, where each rowi (for atomi) is identical to h(b) global. For a batch, this is Hglobal ex∈ RB×N×dp

  81. [92]

    MLP Input Formulation: For each atom i in sample b, the input to the final MLP, MLPblind, is formed by concatenating its corresponding expanded global context and its conditioner vector: M(b,i) in = concat(h(b) global,C (b,i))∈ Rdp+dc Note that h(b) global is the same for all ...

  82. [93]

    Coordinate Prediction: A shared MLP blind processesM(b,i) in for each atom to predict its 3D coordinates: X(b,i) pred = MLPblind(M(b,i) in )∈ R3 This results in the full predicted structureXpred∈ RB×N×3. C.2 Sequential Pooling Strategy The sequential pooling strategy decodes t...

  83. [94]

    Backbone Embedding Selection: Latent embeddings Z(b) bb ∈ RNbb×dz and conditioner vectorsC(b) bb ∈ RNbb×dc corresponding to backbone atomsIbb are selected

  84. [95]

    For a batch, this is Hbb∈ RB×dp,bb

    Backbone Pooling: Z(b) bb is pooled using a 2D adaptive average pooling layer Pbb (e.g., BlindPooling2D from the implementation, with output size Hbb× Wbb) to obtain a backbone-specific context vector: h(b) bb =Pbb(Z(b) bb )∈ Rdp,bb wheredp,bb =Hbb·Wbb. For a batch, this is Hb...

  85. [96]

    C(b) bb is used directly

    Context and Conditioner Expansion: h(b) bb is expanded to H(b) bb ex∈ RNbb×dp,bb. C(b) bb is used directly

  86. [97]

    MLP Input for Backbone Atoms: For each backbone atomj∈I bb in sampleb, the input to the backbone MLP, MLPbb, is: M(b,j) in, bb = concat(h(b) bb ,C (b,j) bb )∈ Rdp,bb+dc

  87. [98]

    Backbone Coordinate Prediction: A shared MLPbb predicts backbone coordinates: Xpred (b,j) bb = MLPbb(M(b,j) in, bb )∈ R3 This yields the predicted backbone structureXpredbb∈ RB×Nbb×3. C.2.2 Sidechain Decoder Stage This stage predicts sidechain atom coordinates Xpredsc using th...

  88. [99]

    Sidechain Embedding Selection: Latent embeddings Z(b) sc ∈ RNsc×dz corresponding to sidechain atomsIsc are selected from the full latent embeddingsZ(b)

  89. [100]

    For a batch, this results in Hsc∈ RB×dp,sc

    Sidechain Pooling: The selected sidechain embeddingsZ(b) sc are pooled using a 2D adap- tive average pooling layer Psc (e.g., BlindPooling2D with output size Hsc×Wsc) to obtain a sidechain-specific context vector for each sample in the batch: h(b) sc =Psc(Z(b) sc )∈ Rdp,sc whe...

  90. [101]

    The construction varies based on the arch type: • Let X(b) pred, bb flat∈ RNbb·3 be the flattened predicted backbone coordinates for sample b

    Feature Construction for Sidechain MLP Input: The input to the sidechain MLP, MLPsc, is a global vector M(b) in, sc constructed per sample b. The construction varies based on the arch type: • Let X(b) pred, bb flat∈ RNbb·3 be the flattened predicted backbone coordinates for sa...

  91. [102]

    Sidechain Coordinate Prediction: The sidechain MLP, MLPsc, processes the constructed input vectorM(b) in, sc (corresponding to the chosen arch type) to predict all sidechain coor- dinates for sampleb simultaneously: X(b) pred, sc flat = MLPsc(M(b) in, sc)∈ RNsc·3 This flattene...

  92. [103]

    Full Structure Assembly: The final predicted structure X(b) pred ∈ RN×3 for sample b is assembled by combining the predicted backbone coordinates X(b) pred, bb and the predicted sidechain coordinatesX(b) pred, sc. C.3 Residue-based Pooling Strategy The Residue-based Pooling st...

  93. [104]

    23 • Pool these embeddings using a 2D adaptive average pooling layer Pres (e.g., nn.AdaptiveAvgPool2d with outputH×W ): h(b) Rj =Pres(Z(b) Rj )∈ Rdp wheredp =H·W

    Per-Residue Pooling: For each residueRj in sampleb: • Select atom embeddings for residue Rj:Z(b) Rj∈ R|Vj|×dz. 23 • Pool these embeddings using a 2D adaptive average pooling layer Pres (e.g., nn.AdaptiveAvgPool2d with outputH×W ): h(b) Rj =Pres(Z(b) Rj )∈ Rdp wheredp =H·W . Th...

  94. [105]

    Atom-Specific Context Assembly: For each atomi in sampleb, its specific context vector is the pooled vector of its parent residueRf(i): h(b,i) atom context = h(b) Rf (i) ∈ Rdp This can be gathered for all atoms to form H(b) atom context∈ RN×dp

  95. [106]

    Conditioner Expansion: The conditionerC∈ RN×dc is expanded for the batch toCex∈ RB×N×dc

  96. [107]

    MLP Input Formulation: For each atomi in sampleb, the input to the final MLP, MLPres, is formed by concatenating its residue’s pooled context and its specific conditioner vector: M(b,i) in, res = concat(h(b,i) atom context,C (b,i))∈ Rdp+dc

  97. [108]

    1level” and “2level

    Coordinate Prediction: A shared MLPres processesM(b,i) in, res for each atom to predict its 3D coordinates: X(b,i) pred = MLPres(M(b,i) in, res)∈ R3 This results in the full predicted structureXpred∈ RB×N×3. This strategy allows the model to learn representations that are loca...

    2024

  98. [3635]

    URL http://dx.doi.org/10.1038/s41392-0 20-00435-w

    doi: 10.1038/s41392-020-00435-w. URL http://dx.doi.org/10.1038/s41392-0 20-00435-w

    doi

  99. [7358]

    URL http://dx.doi.org/10.1371/journal .pcbi.1008936

    doi: 10.1371/journal.pcbi.1008936. URL http://dx.doi.org/10.1371/journal .pcbi.1008936

    doi

Pith tools

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