pith. sign in

arxiv: 2606.22377 · v1 · pith:EVHAAJWMnew · submitted 2026-06-21 · 💻 cs.LG · cs.NA· math.NA

Multigrid Training for Molecular Generation using Graph Neural Networks

Zixuan Ling , Paula Mercurio , Di Liu This is my paper

Pith reviewed 2026-06-26 11:08 UTC · model grok-4.3

classification 💻 cs.LG cs.NAmath.NA
keywords multigrid trainingmolecular generationgraph neural networks3D ligand generationconditional variational autoencoderparameter transfercoarse-to-fine optimizationreceptor-conditioned generation
0
0 comments X

The pith

Multigrid training transfers parameters from coarse molecular graphs and grids to finer ones, accelerating convergence and improving generalization over training from scratch.

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

The paper introduces a multigrid training strategy for models that generate molecular structures as graphs or 3D voxel grids. Low-resolution versions are optimized first, then parameters move to higher resolutions through biased random walk upsampling on graphs or shape-compatible convolutions on grids. This targets the problem that full high-resolution computations scale poorly in cost and stability. A reader would care if the transfer succeeds because it could make receptor-conditioned ligand generation and similar tasks more practical without full recomputation at every scale.

Core claim

The paper claims that pretraining a coarse-resolution conditional variational autoencoder or graph model and transferring parameters to a fine-resolution counterpart via biased random walk upsampling or shape-compatible convolutions accelerates convergence and improves generalization on receptor-conditioned 3D ligand generation compared with training the fine model from scratch.

What carries the argument

Multigrid training strategy that performs low-resolution optimization first then transfers parameters across discretizations using biased random walk upsampling for graphs and shape-compatible convolutions for 3D grids.

If this is right

  • Training time for high-resolution molecular generators decreases because coarse optimization supplies a better starting point.
  • Generalization on receptor-conditioned 3D ligand tasks improves relative to scratch training.
  • The same coarse-to-fine transfer works for both graph neural network representations and voxel-grid CVAEs.
  • Computational cost that normally grows with resolution is partially offset by the staged optimization.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could extend to other grid- or graph-based scientific simulations where resolution scaling is costly.
  • Combining multigrid transfer with existing data augmentation or regularization might yield further gains in stability.
  • Testing transfer across even larger resolution jumps would show where compatibility begins to break.

Load-bearing premise

Parameters learned at coarse resolution stay compatible and useful when moved to finer graphs or grids without needing major extra adaptation or losing key molecular features.

What would settle it

An experiment that trains identical high-resolution models on the same ligand generation task, one from scratch and one initialized via multigrid transfer, then shows the multigrid version converges no faster and generalizes no better.

read the original abstract

Deep learning has demonstrated significant success for modeling biochemical molecular systems, where inputs are commonly represented as graphs or 3D grids. A major challenge is that computational cost scales with resolution, making full graph/grid computation of molecular densities expensive and often unstable. We introduce a multigrid training strategy that leverages low-resolution optimization to accelerate learning at higher resolution through parameter transfer across discretizations. For graph molecular representations, we progressively transfer parameters learned from a coarse graph to a sequence of increasingly finer graphs via biased random walk upsampling. For 3D molecular generation, we voxelize the molecular structures at multiple resolutions, pretrain a coarse-resolution conditional Variational Autoencoder (CVAE), and initialize a fine-resolution CVAE by transferring shape compatible convolutional parameters from the coarse model. Numerical experiments on receptor-conditioned 3D Ligand generation show that multigrid training accelerates convergence and improves generalization compared to training from scratch.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The paper introduces a multigrid training strategy for deep learning models of molecular systems represented as graphs or 3D voxel grids. Parameters learned at coarse resolution are transferred to finer resolutions via biased random walk upsampling (graphs) or shape-compatible convolutions (voxels) in conditional VAEs. The central empirical claim is that this accelerates convergence and improves generalization on receptor-conditioned 3D ligand generation relative to training from scratch.

Significance. If the claimed gains are shown to arise specifically from successful parameter transfer rather than extra compute, the method could meaningfully reduce the cost of high-resolution molecular modeling in drug discovery and related domains. The approach directly targets the resolution-scaling bottleneck in graph- and grid-based biochemical models.

major comments (3)
  1. [Experiments] The central claim rests on successful transfer of parameters learned at coarse resolution, yet the manuscript provides no ablation that isolates the contribution of the transferred parameters from the additional coarse-level pre-training compute budget. Without this isolation (e.g., comparing multigrid initialization against an equivalent extra-epoch baseline at fine resolution), observed speed-ups cannot be attributed to the transfer mechanism.
  2. [Method (parameter transfer subsections)] No quantitative assessment of feature preservation under biased random walk upsampling or shape-compatible convolution transfer is reported (e.g., reconstruction error on held-out molecular densities, graph invariants, or atom-type distributions before versus after upsampling). This leaves the weakest assumption—that critical molecular features survive the transfer—unverified.
  3. [Abstract and Experiments] The abstract states an empirical improvement but supplies no quantitative results, baselines, error bars, dataset sizes, or statistical significance tests. The experiments section must supply these to make the generalization claim verifiable.
minor comments (1)
  1. [Method] Notation for the upsampling operator and the definition of 'shape-compatible' convolutions should be formalized with explicit equations rather than prose descriptions.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and will incorporate revisions to strengthen the empirical support for the multigrid training claims.

read point-by-point responses

  1. Referee: [Experiments] The central claim rests on successful transfer of parameters learned at coarse resolution, yet the manuscript provides no ablation that isolates the contribution of the transferred parameters from the additional coarse-level pre-training compute budget. Without this isolation (e.g., comparing multigrid initialization against an equivalent extra-epoch baseline at fine resolution), observed speed-ups cannot be attributed to the transfer mechanism.

    Authors: We agree that an ablation isolating parameter transfer from extra compute budget is necessary to attribute gains specifically to the transfer. We will add such an experiment in the revised manuscript, comparing multigrid initialization against a fine-resolution baseline trained for an equivalent number of additional epochs without transfer. revision: yes

  2. Referee: [Method (parameter transfer subsections)] No quantitative assessment of feature preservation under biased random walk upsampling or shape-compatible convolution transfer is reported (e.g., reconstruction error on held-out molecular densities, graph invariants, or atom-type distributions before versus after upsampling). This leaves the weakest assumption—that critical molecular features survive the transfer—unverified.

    Authors: We acknowledge this gap in verifying feature preservation during transfer. The revised manuscript will include quantitative metrics, such as reconstruction error on held-out voxel densities and preservation statistics for graph invariants and atom-type distributions, to assess the transfer process. revision: yes

  3. Referee: [Abstract and Experiments] The abstract states an empirical improvement but supplies no quantitative results, baselines, error bars, dataset sizes, or statistical significance tests. The experiments section must supply these to make the generalization claim verifiable.

    Authors: We will revise the abstract to report key quantitative results including convergence improvements, generalization metrics with error bars, and statistical tests. The experiments section will be expanded to include dataset sizes, complete baselines, and significance testing to make the claims verifiable. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical validation of multigrid transfer method

full rationale

The paper introduces a multigrid training strategy for graph and voxel-based molecular generation, with parameter transfer via biased random walk upsampling or shape-compatible convolutions. The central claim (accelerated convergence and improved generalization on receptor-conditioned 3D ligand generation) is presented strictly as an empirical result from numerical experiments, not as a mathematical derivation or prediction. No equations, fitted parameters renamed as predictions, self-definitional constructs, or load-bearing self-citations appear in the provided text. The method's assumptions about parameter compatibility are acknowledged as empirical and unproven by ablation in the skeptic notes, but this is a correctness/evidence issue, not circularity. The derivation chain is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The method rests on the domain assumption that coarse-resolution parameters transfer usefully to fine resolutions; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Parameters optimized at coarse molecular discretizations remain compatible and beneficial when transferred to finer discretizations
    This premise is required for the parameter-transfer step described in the abstract to accelerate rather than degrade training.

pith-pipeline@v0.9.1-grok · 5685 in / 1216 out tokens · 24824 ms · 2026-06-26T11:08:51.920469+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 22 canonical work pages · 2 internal anchors

  1. [1]

    AHMED, R

    A. AHMED, R. D. SMITH, J. J. CLARK, J. DuNBAR, James B., AND H. A. Carlson, Recent im- rovements to binding moad: a resource for protein-ligand binding affinities and structures Nucleic Acids Research, 43(2015), pp. D465-D469, https://doi.org/10.1093/nar/gku1088

    work page doi:10.1093/nar/gku1088 2015

  2. [2]

    H. M. BERMAN, J. WESTBROOK, Z. FENG, G. GILLILAND, T. N. BHAT, H. WEISSIG, I. N. pp. 235-242, https://doi.org/10.1093/nar/28.1.235. 3 G. R. BIckeRtoN, G. V. PAOLINI, J. BESNARD, S. MURESAN, AND A. L. HopkIns, Quantifying 1038/nchem.1243. A. BRANDT, Multi-level adaptive solutions to boundary-value problems, Mathematics of Con itation, 31 (1977), pp. 333-39...

    work page doi:10.1093/nar/28.1.235 1977

  3. [3]

    N. D. CAO AND T. KIPF, Molgan: An implicit generative model for small molecular graphs, arXiv preprint arXiv:1805.11973, (2018), https://doi.org/10.48550/arXiv.1805.11973

    work page doi:10.48550/arxiv.1805.11973 2018

  4. [4]

    CHILD, Very deep vaes, arXiv preprint arXiv:2011.10650, (2020), https://doi.org/10.48550/ arXiv.2011.10650

    R. CHILD, Very deep vaes, arXiv preprint arXiv:2011.10650, (2020), https://doi.org/10.48550/ arXiv.2011.10650. //openreview.net/forum?id=SyqShMZRb

    arXiv 2011

  5. [5]

    C. M. DoBson, Chemical space and biology, Nature, 432(2004), pp. 824-828, https://doi.org/ 10.1038/nature03192. MULTIGRID GRAPH NEURAL NETWORKS 27

    work page doi:10.1038/nature03192 2004

  6. [6]

    ERTL AND A

    P. ERTL AND A. SCHUFFENHAUER, Estimation of synthetic accessibility score of drug-like mol- ecules based on molecular complexity and fragment contributions, Journal of Cheminfor- matics, 1 (2009), p. 8, https://doi.org/10.1186/1758-2946-1-8. pp. 4200-4215, https://doi.org/10.1021/acs.jcim.0c00411

    work page doi:10.1186/1758-2946-1-8 2009

  7. [7]

    GAO AND S

    H. GAO AND S. JI, Graph u-nets, in Proceedings of the 36th International Conference on Machine Learning (ICML 2019), vol. 97 of Proceedings of Machine Learning Research, PMLR, 2019, pp. 2083-2092, https://proceedings.mlr.press/v97/gao19a.html

    2019

  8. [8]

    GAULTON, L

    A. GAULTON, L. J. BELLIS, A. P. BENTO, J. CHAMBERS, M. DAVIES, A. HERSEY, Y. LIGHT, S. McGLINCHEY, D. MICHALOVICH, B. AL-LAZIKANI, AND J. P. OVERINGTON, Chembl: a large-scale bioactivity database for drug discovery, Nucleic Acids Research, 40 (2012), pp. D1100-D1107, https://doi.org/10.1093/nar/gkr777

    work page doi:10.1093/nar/gkr777 2012

  9. [9]

    GóMEZ-BOMBARELLI, J

    R. GóMEZ-BOMBARELLI, J. N. WEI, D. DUVENAUD, J. M. HERNÁNDEZ-LOBATO, B. SÁNCHEZ- LENGELING, D. SHEBERLA, J. AGUILERA-IPARRAGUIRRE, T. D. HIRZEL, R. P. ADAMS, AND A. AsPURU-GuzIK, Automatic chemical design using a data-driven continuous representa- tion of molecules, ACS Central Science, 4 (2018), pp. 268-276, https://doi.org/10.1021/ acscentsci.7b00572. P...

    2018

  10. [10]

    KDD - node2vec: Scalable Feature Learning for Networks,

    A. GROVER AND J. LESKOVEC, node2vec: Scalable feature learning for networks, in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD 2016), ACM, 2016, pp. 855-864, https://doi.org/10.1145/2939672.2939754, https://doi.org/10.1145/2939672.2939754

    work page doi:10.1145/2939672.2939754 2016

  11. [11]

    J. GUAN, W. W. QIAN, X. PENG,Y. Su, J. PENG, AND J. MA, for target-aware molecule generation and affinity prediction, 2023, https://arxiv.org/abs/ 2303.03543, https://arxiv.org/abs/2303.03543. [16 J. GUAN, X. ZHOU, Y. YANG, Y. BAO, J. PENg, J. Ma, Q. LIu, L. WANG, AND Q. Gu, De- compdiff: Diffusion models with decomposed priors for structure-based drug des...

    arXiv 2023

  12. [12]

    W. L. HAMILTON, R. YING, AND J. LESKOVEC, Inductive representation learning on large graphs, in Proceedings of the 31st International Conference on Neural Information Process- ing Systems, NIPS'17, Red Hook, NY, USA, 2017, Curran Associates Inc., pp. 1025-1035. 18] J. HocHULI, A. HELBLING, T. SKAIST, M. RAGOZA, AND D. R. Koes, Visualizing convolutional ne...

    work page doi:10.1016/j.jmgm.2018.06.005 2017

  13. [13]

    W. Hu, B. LIU, J. GoMES, M. ZITNIK, P. LIANG, V. PANDE, AND J. LESKOVEC, Strategies for pre-training Graph Neural Networks, in International Conference on Learning Representa- tions (ICLR), 2020, https://dblp.org/rec/conf/iclr/HuLGZLPL20.html. 21] W. JIN, R. BARZILAY, AND T. JAAKKOLA, Junction tree variational autoencoder for molec- ular graph generation,...

    work page doi:10.1039/9781788016841-00228 2020

  14. [14]

    JUMPER, R

    J. JUMPER, R. EVANS, A. PRITZEL, T. GREEN, M. FIGURNOV, O. RoNNEBERGEr, K. TUN- YASUVUNAKOOL, R. BATES, A. ZÍDEK, A. POTAPENKO, A. BRIDGLAND, C. MEYER, S. A. A. KoHL, A. J. BALLARD, A. COWIE, B. ROMERA-PAREDES, S. NIKOLOV, R. JAIN, J. ADLER, T. BACK, S. PETERSEN, D. REIMAN, E. CLANCY, M. ZIELINSKI, M. STEINEG- GER, M. PACHOLSKA, T. BERGHAMMER, S. BODENSTE...

    2021

  15. [15]

    T. N. KIPF AND M. WELLING, Semi-supervised classification with graph convolutional net- works, in International Conference on Learning Representations (ICLR), 2017, https: //openreview.net/forum?id=SJU4ayYgl. arXiv:1609.02907

    Pith/arXiv arXiv 2017

  16. [16]

    M. J. KUSNER, B. PAIGE, AND J. M. HERNÁNDEZ-LOBATO, Grammar variational autoencoder, n Proceedings of the 34th International Conference on Machine Learning, D. Precup anc . W. Teh, eds., vol. 70 of Proceedings of Machine Learning Research, PMLR, 2017 pp. 1945-1954, https://proceedings.mlr.press/v70/kusner17a.html. 28 ZIXUAN LING, DI LIU, PAULA MERCURIO 25...

    2017

  17. [17]

    M. LIu, Y. Luo, K. UcHINo, K. MARUHASHI, AND S. JI, Generating 3D molecules for target pro- tern binding, 2022, https://arxiv.org/abs/2204.09410, https://arxiv.org/abs/2204.09410. (27] T. LIU, L. HWANG, S. K. BURLEY,C. I. NITSCHE, C. SOUTHAN, W. P. WALTERS, AND M. K. GIlson, Bindingdb in 2024: a fair knowledgebase of protein-small molecule binding data, N...

    work page doi:10.1093/nar/ 2022

  18. [18]

    Z. LIu, Y. LI, L. HAN, J. Li, J. Liu, Z. ZHAo, W. NIE, Y. LIU, AND R. WANg, Pdb-wide col- lection of binding data: current status of the pdbbind database, Bioinformatics, 31 (2015), pp. 405-412, https://doi.org/10.1093/bioinformatics/btu626. 29] S. Luo, J. GuAN, J. MA, AND J. PENG, A 3D generative model for structure-based drug design, 2022, https://arxiv...

    work page doi:10.1093/bioinformatics/btu626 2015

  19. [19]

    MARSDEN, N

    A. MARSDEN, N. MoKRZECKI, E. De GIORGI, L. VoST, AND C. M. DEANE, Molsnapper: Conditioning diffusion for structure-based drug design, J. Chem. Inf. Model., 65 (2025), pp. 4263-4273, https://doi.org/10.1021/acs.jcim.4c02008, http://dx.doi.org/10.1021/acs

    work page doi:10.1021/acs.jcim.4c02008 2025

  20. [20]

    A. T. McNUTT, P. FRANCOEUR, R. AGGARWAL, T. MASUDA, R. MELI, M. RAGOZA, J. SUNSERI, AND D. R. KOES, Gnina 1.0: molecular docking with deep learning, Journal of Cheminfor- matics, 13 (2021), p. 43, https://doi.org/10.1186/s13321-021-00522-2.33] M. OLIVECRONA, T. BLASCHKE, O. ENGKVIST, AND H. CHEN, Molecular de-novo design through deep reinforcement learnin...

    work page doi:10.1186/s13321-021-00522-2.33 2021

  21. [21]

    X. PENG, J. GUAN, Q. LIu, AND J. Ma, MolDiff: Addressing the atom-bond inconsistency problem in 3D molecule diffusion generation, in Proceedings of the 40th International Conference on Machine Learning, vol. 202 of Proceedings of Machine Learning Research, MLR, 2023, pp. 27611-27629, https://proceedings.mlr.press/v202/peng23b.html. 35] X. PeNg, S. Luo, J....

    work page doi:10.1021/acs.jcim.6b00740 2023

  22. [22]

    Ligand Pose Optimization with Atomic Grid-Based Convolutional Neural Networks

    M. RAGOZA, T. MASUDA, AND D. R. KoEs, Generating 3d molecules conditional on receptor binding sites with deep generative models, Chem. Sci., 13 (2022), pp. 2701-2713, https: 39] M. RAGOZA, L. TURNER, AND D. R. KoEs, Ligand pose optimization with atomic grid-based convolutional neural networks, 2017, https://doi.org/10.48550/arXiv.1710.07400, https:// doi....

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1710.07400 2022

  23. [23]

    URL https://doi.org/10.1021/ci300415d

    L. RUDDIGKEIT, R. VAN DEURSEN, L. C. BLUM, AND J. REYMOND, Enumeration of 166 billion organic small molecules in the chemical universe database gdb-17, Journal of Chemical Information and Modeling, 52(2012), pp. 2864-2875, https://doi.org/10.1021/ci300415d

    work page doi:10.1021/ci300415d 2012

  24. [24]

    Nature Computational Science4(12), 899–909 (2024) https://doi.org/10.1038/s43588-024-00737-x 36

    A. SCHNEVING, C. HARRIS, Y. Du, K. DIDI, A. JAMASB, I. IGASHOV, W. Du, C. GoMES, T. L. BLUNDELL, P. LIO, M. WELLING, M. BRONSTEIN, AND B. CORREIA, Structure-based drug design with equivariant diffusion models, Nat. Comput. Sci., 4(2024), pp. 899-909, https: //doi.org/10.1038/s43588-024-00737-x, http://dx.doi.org/10.1038/s43588-024-00737-x. MULTIGRID GRAPH...

    work page doi:10.1038/s43588-024-00737-x 2024

  25. [25]

    SIMONOVSKY AND N

    M. SIMONOVSKY AND N. KOMODAKIS, Graphvae: towards generation of small graphs using variational autoencoders, in Artificial neural networks and machine learning - icann 2018, 2018, Springer International Publishing, pp. 412-422, https://doi.org/10.1007/

    2018

  26. [26]

    C. K. SØNDERBY, T. RAIKO, L. MAALØE, S. K. SØNDERBY, AND O. WINTHER, Ladder vari- ational autoencoders, arXiv preprint arXiv:1602.02282, (2016), https://doi.org/10.48550/

    Pith/arXiv arXiv 2016

  27. [27]

    J. E. SUNSERI, J. E. KING, P. G. FRANCOEUR, AND D. R. KoEs, Convolutional neural network Aided Molecular Design, 33 (2019), pp. 19-34, https://doi.org/10.1007/s10822-018-0133-y, https://doi.org/10.1007/s10822-018-0133-y.47] A. VAHDAT AND J. KAUTZ, Nvae: A deep hierarchical variational autoencoder, in Advances in Neural Information Processing Systems (Neur...

    work page doi:10.1007/s10822-018-0133-y 2019

  28. [28]

    L. VosT, Y. ZIv, AND C. M. DEANE, Incorporating targeted protein structure in deep learning methods for molecule generation in computational drug design, Chem. Sci., 16 (2025), pp. 20677-20693, https://doi.org/10.1039/D5SC05748E, https://doi.org/10.1039/ D5SC05748E. 50] I. WALLACH, M. DZAMBA, AND A. HEIFETS, Atomnet: A deep convolutional neural network fo...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1039/d5sc05748e 2025

  29. [29]

    M. J. WARING, J. ARRoWsMITH, A. R. LEACH, P. D. LEEsON, S. MANDRELL, R. M. OWEN, G. PAIRAUDEAU, W. D. PENNIE, S. D. PICKETT, J. WANG, O. WALLACE, AND A. WEIR, An analysis of the attrition of drug candidates from four major pharmaceutical companies, Nature Reviews Drug Discovery, 14(2015), pp. 475-486, https://doi.org/10.1038/nrd4609

    work page doi:10.1038/nrd4609 2015

  30. [30]

    J. A. WELLER AND R. Rohs, Structure-based drug design with a deep hierarchical generative model, Journal of Chemical Information and Modeling, 64(2024), pp. 6450-6463, https: //doi.org/10.1021/acs.jcim.4c01193. PMID: 39058534

    work page doi:10.1021/acs.jcim.4c01193 2024

  31. [31]

    WEN AND D

    Z. WEN AND D. GOLDFARB, A line search multigrid method for large-scale nonlinear optimiza- tion, SIAM Journal on Optimization, 20 (2009), pp. 1478-1503, https://doi.org/10.1137/ 08071524X

    2009

  32. [32]

    Z. Wu, B. RAMSUNDAR, E. N. FEINBERG, J. GOMES, C. GENIESSE, A. S. PAPPU, K. LESWING, AND V. PANDE, Moleculenet: a benchmark for molecular machine learning, Chemical Sci- ence, 9 (2018), pp. 513-530, https://doi.org/10.1039/C7SC02664A. 55 K. Xu, W. HU, J. LESKOVEC, AND S. JEGELKA, How powerful are Graph Neural Networks?, in International Conference on Lear...

    work page doi:10.1039/c7sc02664a 2018

  33. [33]

    Z. YINg, J. You, C. MoRRIs, X. REN, W. L. HAMILTON, AND J. LESKOVEC, Hierarchical graph representation learning with differentiable pooling, in Advances in Neural Information Processing Systems 31 (NeurIPS 2018), 2018, pp. 4805-4815, https://proceedings.neurips. cc/paper/2018/hash/e77dbaf6759253c7c6d0efc5690369c7-Abstract.html

    2018

  34. [34]

    J. You, Z. LIu, L. SONG, ET AL., Graph convolutional policy network for goal-directed molec- ular graph generation, arXiv preprint arXiv:1806.02473, (2018), https://doi.org/10.48550/ arXiv.1806.02473

    Pith/arXiv arXiv 2018

  35. [35]

    ZHUNG, H

    W. ZHUNG, H. KIM, AND W. Y. KIm, 3d molecular generative framework for interaction- guided drug design, Nature Communications, 15 (2024), p. 2688, https://doi.org/10.1038/ s41467-024-47011-2, https://www.nature.com/articles/s41467-024-47011-2

    2024