Pith. sign in

REVIEW 3 major objections 5 minor 73 references

Generating 3D Binding Molecules Using Shape-Conditioned Diffusion Models with Guidance

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read DiffSMol generates novel 3D drug-like molecules conditioned on the shapes of known ligands, reporting a 61.4% shape-similar success rate versus 11.2% for the best baseline, and pocket guidance that improves predicted binding affinity by…

desk verdict The SMG story is solid; the PMG headline numbers fail an equal-sample comparison. read the letter →

arxiv 2502.06027 v1 pith:EIJ52L5V submitted 2025-02-09 cs.LG

Ziqi Chen , Bo Peng , Tianhua Zhai , Daniel Adu-Ampratwum , Xia Ning This is my paper · ORCID
classification cs.LG
keywords 3Dmoleculegenerationdiffusionmodelsshape-conditionedequivariantgraphneuralnetworksligand-baseddrugdesignstructure-basedbindingaffinityshapeembeddings
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 sets out to show that a generative model can turn the 3D surface shape of a known ligand into a family of novel, drug-like binding molecules, and that a diffusion model conditioned on learned shape embeddings does this better than existing shape-conditioned and pocket-conditioned generators. On the shape-matching benchmark, DiffSMol with shape guidance produces desirable molecules (high shape similarity, low graph similarity) at a 61.4% rate, versus 11.2% for the best baseline, while 99.8-99.9% of its desirable molecules are novel in graph structure. Trained only on molecule data and never on protein-ligand complexes, the same model with pocket guidance improves predicted binding affinity over the best pocket-conditioned baseline by 13.2%, or 17.7% when shape guidance is added. The reason to care: generating molecules that resemble known binders while differing chemically is a direct, automatable route to new drug candidates, and this is one of the first systems to unify the usually separate toolchains of ligand-based and structure-based design.

What carries the argument

The load-bearing object is the pre-trained equivariant shape embedding $H_s$: the module SE encodes a ligand's molecular surface, sampled as a point cloud, into a rotation-equivariant latent embedding by training a decoder to recover the signed distances of query points to the surface. That embedding conditions a second module, DIFF, an equivariant denoising diffusion model whose prediction network SMP interleaves geometric vector perceptrons with shape-aware atom representation layers that inject $H_s$ at every layer, plus bond-type prediction as an auxiliary training signal. At inference, two guidance mechanisms carry the performance: shape guidance (SG) moves predicted atom positions toward points sampled around the condition molecule's atoms, applied only in the early high-noise steps, and pocket guidance (PG) repels molecule atoms from protein atoms closer than a learned threshold. The shape guidance is the difference between a 28.4% and a 61.4% desirable-molecule rate.

What would settle it

Synthesize a random sample of the molecules DiffSMol marks as desirable ($S_{ims} \ge 0.8$ with low graph similarity) for a chosen target and measure their binding by surface plasmon resonance or isothermal titration calorimetry; if a large fraction show no measurable binding despite the high shape similarity, the proxy on which the success-rate claim rests is falsified. A cheaper first check is retrospective: on a set of ligands with known binding data, test whether the $S_{ims} \ge 0.8$ threshold actually separates binders from non-binders.

Watch

Extended reading notes

Core claim

The central claim, stated as the authors would state it to a fair reader, is that DiffSMol captures the geometric essence of a ligand's shape in pre-trained equivariant embeddings and then diffuses atom types and positions conditioned on those embeddings, yielding molecules that are at once shape-similar to the condition ligand, novel in molecular graph, realistic in 3D structure, and drug-like. Adding shape guidance, which pushes predicted atom positions toward the condition molecule's shape during the noisy early denoising steps, raises the rate of desirable molecules ($S_{ims} \ge 0.8$ with graph dissimilarity below a threshold) from 28.4% to 61.4% at the strictest threshold, far above the 11.2% best baseline, without degrading novelty or diversity. With protein-pocket guidance, which repels generated atoms from protein atoms that come too close, DiffSMol achieves better AutoDock Vina binding-affinity scores than the best pocket-conditioned baseline by 13.2%, and by 17.7% when shape guidance is also active, despite never being trained on protein-ligand complexes. Case studies on CDK6 and neprilysin report generated molecules whose predicted binding affinities beat the known co-crystallized ligands and whose physicochemical, toxicity, and ADMET profiles compare with approved drugs.

Load-bearing premise

The load-bearing premise is that ROCS shape overlap ($S_{ims} \ge 0.8$) and AutoDock Vina affinity scores are trustworthy proxies for real binding activity; the entire evaluation, including the 61.4% success rate, rests on that proxy, and the paper states that no wet-lab binding validation was performed.

Editorial extensions

If this is right

  • Shape-conditioned generation does not have to sacrifice novelty: 99.8-99.9% of DiffSMol's desirable molecules are absent from the MOSES training set, while its shape similarity matches or exceeds the baselines.
  • Molecule-only pretraining can substitute for scarce protein-ligand complex data: DiffSMol is never trained on complexes, yet with pocket guidance it beats complex-trained pocket-conditioned baselines on predicted binding affinity.
  • Shape and pocket guidance are complementary: adding shape guidance on top of pocket guidance raises the binding-affinity improvement over the best baseline from 13.2% to 17.7%.
  • Generation is fast enough for large candidate screens: DiffSMol produces 100 molecules in 48-58 seconds versus 1,252 seconds for the fastest pocket-conditioned baseline, enabling more than ten times the candidate pool in the same time.

Reading between the lines

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

  • The pocket-conditioned comparison allocates different effort per method: the paper has DiffSMol generate 1,000 molecules per pocket and select its best 100 by Vina/QED/SA scores, while baselines generate 100 molecules with no post-hoc selection; a matched-budget rerun would clarify how much of the reported gap is selection budget rather than generation quality.
  • The shape-guidance step nudges predicted atom positions toward points sampled around the condition molecule's own atoms, which amounts to geometry copying at the atom level; dissecting which atoms the guidance actually displaces would reveal whether it enforces a pharmacophore-like arrangement or largely memorizes coordinates.
  • The authors note that pocket guidance uses only geometry; a natural extension their architecture already hints at is conditioning on pocket electrostatics and hydrophobicity, and multi-objective guidance that optimizes ADMET properties alongside shape and affinity.
  • Because the evaluation is entirely in silico, the decisive follow-up is to synthesize a sample of the desirable molecules and measure binding in vitro; the architecture's speed (under a minute for 100 molecules) makes such a screen feasible at scale.
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

3 major / 5 minor

Summary. The manuscript introduces DiffSMol, a 3D molecule generation method conditioned on ligand shape embeddings learned by a pre-trained equivariant shape encoder, with an optional inference-time shape guidance step and an optional pocket guidance step for structure-based generation. The method is evaluated on two tasks: shape-conditioned molecule generation (SMG) on the MOSES dataset against SQUID and virtual screening, and pocket-conditioned molecule generation (PMG) on CrossDocked2020 against AR, Pocket2Mol, TargetDiff, and DecompDiff. The main reported results are a 61.4% success rate for DiffSMol with shape guidance on SMG versus an 11.2% best baseline, and a 13.2%/17.7% improvement in binding affinity for pocket-guided variants on PMG, together with case studies for CDK6 and NEP.

Significance. The SMG results are internally consistent and, if taken at face value, represent a substantial advance: Table 2 shows DiffSMol reaches 28.4% desired-molecule rate without guidance and 61.4% with guidance, versus 11.2% for the best baseline, with high novelty and diversity. The method ships with public code and data, and the diffusion formulation follows standard, well-documented practice. However, the abstract's PMG binding-affinity claims are not supported by the paper's own equal-sample comparison: Table S9 shows that when all methods generate 100 molecules, DiffSMol+p and DiffSMol+s+p have worse average Vina S than AR and worse Vina D than TargetDiff. The PMG headline improvement is an artifact of selecting the top 100 of 1,000 samples by the evaluation metrics themselves. The central SMG contribution is therefore viable, while the PMG claims need to be reworked or substantially qualified.

major comments (3)
  1. [Overall Comparison for PMG; Table 4; Table S9] The PMG comparison is not symmetric. The text states explicitly: "For DiffSMol, we generate 1,000 molecules and select the top 100 molecules for comparison based on their Vina S, QED, and SA scores," while each baseline generates 100 molecules with no such selection. Selecting on the evaluation metrics, including Vina S itself, inflates the reported binding affinities and makes the "13.2%" and "17.7%" improvements in the abstract and Section "Overall Comparison for PMG" an artifact of sampling effort rather than model quality. The paper's own Supplementary Table S9, where all methods generate 100 molecules, shows DiffSMol+p and DiffSMol+s+p with average Vina S of -4.15 and -4.56 kcal/mol, both worse than AR (-5.06 kcal/mol), and Vina D of -6.49 and -6.60 kcal/mol, both worse than TargetDiff (-7.37 kcal/mol). To support the claimed binding-affinity superiority, the equal-sample comparison must be the primary evidence, or the baselines must be given the same selection procedure; otherwise the PMG claims should be withdrawn or revised.
  2. [Overall Comparison for PMG; Table 4] The improvement percentages in the text do not match Table 4. The text says DiffSMol+p and DiffSMol+s+p show "17.7% and 13.2% improvement over the best baseline AR (-5.06 kcal/mol)," but from Table 4, DiffSMol+p has Vina S -5.53, which is approximately 9.3% better than -5.06, and DiffSMol+s+p has -5.81, which is approximately 14.8% better than -5.06. The labels appear to be swapped, and the percentages are not correct under any standard definition of improvement on negative binding energies. The authors should recompute and clearly define how percentage improvement is calculated for negative-valued scores.
  3. [Supplementary Section S4; Table S9] The equal-sample Table S9 should be treated as the main PMG result rather than a supplementary caveat. Under this comparison, DiffSMol+p and DiffSMol+s+p still show competitive QED (0.67 and 0.66 vs. 0.58 for Pocket2Mol) and HA (58.52% and 58.28% vs. 57.57% for TargetDiff), but the binding-affinity advantage in Vina S and Vina D disappears. The manuscript should present these results in the main text and temper the abstract and conclusion accordingly, or otherwise justify why selection on the evaluation metric is an acceptable protocol.
minor comments (5)
  1. [Guidance-induced Inference; Fig. 6] The text says shape guidance is "shown in Figure 6(c)" and pocket guidance in "Figure 6(e)", but in the caption Figure 6(c) is pocket guidance and Figure 6(e) is shape guidance; these cross-references should be corrected.
  2. [Abstract and throughout] The method name is spelled inconsistently as "DiffSmol" in the abstract and "DiffSMol" in the body; the spelling should be unified.
  3. [Case Studies for Targets] The Vina S score reported for the 4AU ligand in the CDK6 case study is +0.736 kcal/mol, a positive value that is unusual for a bound ligand and inconsistent with the negative Vina M (-5.939) and Vina D (-7.592) reported for the same ligand; the authors should verify this value and explain why Vina S is positive.
  4. [Experimental Setup; Overall Comparison for PMG] The statement "following Long et al., for baselines, we apply them to generate 100 molecules" cites an SMG paper for a PMG protocol; this should be replaced with the appropriate PMG references or with the original experimental settings of AR, Pocket2Mol, TargetDiff, and DecompDiff.
  5. [DiffSMol with Shape Guidance; Eq. 35; Supplementary S1.2] The shape guidance mechanism introduces several free parameters (sigma, gamma, stop step S, rho, epsilon) that are tuned on the validation set. A brief sensitivity analysis or ablation over these parameters would help establish how robust the headline SMG and PMG results are to their values.

Circularity Check

1 steps flagged · score 6.0 of 10

PMG binding-affinity claim is an artifact of selecting top-100 molecules by the reported Vina S; Table S9 equal-sample comparison removes the advantage.

  1. fitted input called prediction [Overall Comparison for PMG, Table 4 paragraph; Supplementary Section S4, Table S9]
    "Therefore, following Long et al.,21 for baselines, we apply them to generate 100 molecules for each test protein target. For DiffSMol, we generate 1,000 molecules and select the top 100 molecules for comparison based on their Vina S, QED, and SA scores."

    The reported PMG binding-affinity improvement is produced by the evaluation protocol itself: DiffSMol's 100 evaluated molecules are chosen as the best 100 of 1,000 with respect to Vina S (plus QED and SA), while baselines contribute an unselected 100-molecule set. Because Vina S is both a selection criterion and the headline reported metric, the 'prediction' of higher Vina S is statistically forced by the selection step rather than by the generative model. The paper's own equal-sample Table S9 confirms this: DiffSMol+p and DiffSMol+s+p achieve average Vina S of -4.15 and -4.56, both worse than AR (-5.06), so the abstract's 13.2% and 17.7% improvements are an artifact of sampling/selection effort, not model quality.

full rationale

DiffSMol's generative core is standard: the diffusion forward/backward processes (Eqs. 6-18) follow Ho et al. and Hoogeboom et al., and SMP is trained with position, feature, and bond losses (Eqs. 19-23). The pre-trained shape encoder SE is learned from signed-distance reconstruction (Eq. 5) and is not defined in terms of the downstream evaluation metrics. The SMG comparison (Table 2) is an external benchmark against SQUID and VS on the MOSES split, and the reported shape-similarity gains are not a mathematical consequence of Eq. 35 alone, so the SMG claim is not circular. The one substantive circularity is in the PMG evaluation: the paper's headline binding-affinity improvements come from a protocol in which DiffSMol generates 1,000 molecules and selects the top 100 by Vina S, QED, and SA, while baselines contribute 100 unselected molecules. The reported metric (Vina S) is itself one of the selection criteria, so the improvement is partly constructed rather than measured. The paper's own equal-sample Table S9 shows the advantage disappears (DiffSMol+p Vina S -4.15 and DiffSMol+s+p -4.56 vs AR -5.06), confirming that the main-text claim reduces to the selection rule. No self-citation is load-bearing, and no uniqueness theorem or ansatz is smuggled in via citation.

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

The central claims rest on a small number of hand-chosen guidance hyperparameters (S, gamma, sigma, rho, epsilon), on standard diffusion mathematics, and on domain assumptions that shape similarity and in silico scores are valid proxies for binding activity. No new physical entities are introduced.

free parameters (6)
  • Shape guidance stop step S (Eq. 36) = 300
    Hand-chosen; stops shape guidance after 300 of 1000 diffusion steps, affecting shape similarity and validity.
  • Shape guidance distance threshold gamma (Eq. 35) = 0.2
    Hand-chosen threshold that decides when predicted atoms are pushed toward the condition molecule's atom-centered Gaussian cloud.
  • Shape guidance balance sigma (Eq. 35) = sampled from [0.2, 0.8]
    Randomly sampled at each guided update; controls how strongly generated atom positions are pulled to the condition shape.
  • Pocket guidance clash threshold rho (Eq. 37) = identified from training complexes
    Determines the distance below which generated atoms are pushed away from protein atoms; fitted to training complexes.
  • Pocket guidance margin epsilon (Eq. 37) = sampled from [0, 0.5]
    Adds an extra margin to the clash correction; randomly sampled during inference.
  • Loss weight xi and step-weight threshold delta = xi=100, delta=10
    Hand-chosen to balance coordinate, atom type, and bond type losses; affects training and generated molecule quality.
assumptions (5)
  • domain assumption Molecules with similar 3D shapes tend to have similar binding activities (Bostrom et al.).
    This is the motivation for shape-conditioned generation and the validity of using shape similarity as the primary success metric in SMG; stated in the Introduction and Evaluation Metrics.
  • standard math Standard Gaussian and categorical diffusion posteriors from Ho et al. and Hoogeboom et al. apply to 3D atom coordinates and atom types.
    Equations 6-15 and Supplementary Section S10 rely on these published derivations.
  • standard math Equivariance and invariance of Vector Neuron, DGCNN, and GVP modules hold as described in refs 58 and 64.
    The architecture claims rotation equivariance by construction; the paper does not re-derive these guarantees.
  • domain assumption AutoDock Vina scores and in silico ADMET predictions are reliable proxies for binding affinity and drug properties.
    All PMG evaluation and case-study claims use Vina and admetSAR/ICM predictions rather than experimental measurements; the Discussion acknowledges in vitro validation is outside scope.
  • domain assumption Bond types can be determined post hoc from atom types and atomic distances (following refs 26 and 27).
    Used during inference to construct the final molecular graph; stated in Guidance-induced Inference.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generating 3D Binding Molecules Using Shape-Conditioned Diffusion Models with Guidance." pith.science (2026). https://pith.science/paper/EIJ52L5V

@misc{pith2026250206027,
  author       = {Pith},
  title        = {Pith review of: Generating 3D Binding Molecules Using Shape-Conditioned Diffusion Models with Guidance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EIJ52L5V}},
  note         = {Machine review of arXiv:2502.06027}
}
read the original abstract

Drug development is a critical but notoriously resource- and time-consuming process. In this manuscript, we develop a novel generative artificial intelligence (genAI) method DiffSMol to facilitate drug development. DiffSmol generates 3D binding molecules based on the shapes of known ligands. DiffSMol encapsulates geometric details of ligand shapes within pre-trained, expressive shape embeddings and then generates new binding molecules through a diffusion model. DiffSMol further modifies the generated 3D structures iteratively via shape guidance to better resemble the ligand shapes. It also tailors the generated molecules toward optimal binding affinities under the guidance of protein pockets. Here, we show that DiffSMol outperforms the state-of-the-art methods on benchmark datasets. When generating binding molecules resembling ligand shapes, DiffSMol with shape guidance achieves a success rate 61.4%, substantially outperforming the best baseline (11.2%), meanwhile producing molecules with novel molecular graph structures. DiffSMol with pocket guidance also outperforms the best baseline in binding affinities by 13.2%, and even by 17.7% when combined with shape guidance. Case studies for two critical drug targets demonstrate very favorable physicochemical and pharmacokinetic properties of the generated molecules, thus, the potential of DiffSMol in developing promising drug candidates.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 66 canonical work pages

  1. [1]

    & Zhou, S

    Sun, D., Gao, W., Hu, H. & Zhou, S. Why 90% of clinical drug development fails and how to improve it? Acta Pharm. Sin. B. 12, 3049–3062 (2022)

    work page 2022

  2. [2]

    J., McKee, M

    Wouters, O. J., McKee, M. & Luyten, J. Estimated research and development investment needed to bring a new medicine to market, 2009-2018. JAMA 323, 844 (2020)

    work page 2020

  3. [3]

    & MacKerell, A

    Yu, W. & MacKerell, A. D. Computer-Aided Drug Design Methods, 85–106 (Springer New York, 2016)

    work page 2016

  4. [4]

    Acharya, C., Coop, A., Polli, J. E. & MacKerell, A. D. Recent advances in ligand-based drug design: Relevance and utility of the conformationally sampled pharmacophore approach. Curr. Comput. Aided Drug Des.7, 10–22 (2011)

    work page 2011

  5. [5]

    Anderson, A. C. The process of structure-based drug design. Chem. Biol. 10, 787–797 (2003)

    work page 2003

  6. [6]

    Gimeno, A. et al. The light and dark sides of virtual screening: What is there to know? Int. J. Mol. Sci.20, 1375 (2019)

    work page 2019

  7. [7]

    Kingma, D. P. & Welling, M. Auto-encoding variational bayes. arXiv:1312.6114 (2013)

    arXiv 2013

  8. [8]

    & Ermon, S

    Song, J., Meng, C. & Ermon, S. Denoising diffusion implicit models. In 9th International Conference on Learning Representations (2021)

    work page 2021

Show all 73 references
  1. [9]

    GPT-4 technical report

    OpenAI et al. GPT-4 technical report. arXiv:2303.08774 (2023)

    2023 arXiv

  2. [10]

    N., Chen, Z., Ning, X

    Yu, B., Baker, F. N., Chen, Z., Ning, X. & Sun, H. Llasmol: Advancing large language models for chemistry with a large-scale, comprehensive, high-quality instruction tuning dataset. arXiv:2402.09391 (2024)

    2024 arXiv

  3. [11]

    & Jaakkola, T

    Jin, W., Barzilay, R. & Jaakkola, T. Junction tree variational autoencoder for molecular graph generation. In Dy, J. & Krause, A. (eds.) Proceedings of the 35th International Conference on Machine Learning, vol. 80 of Proceedings of Machine Learning Research, 2323–2332 (PMLR, 2018)

    2018

  4. [12]

    Schneuing, A. et al. Structure-based drug design with equivariant diffusion models. arXiv:2210.13695 (2022)

    2022 arXiv

  5. [13]

    Liu, S. et al. Conversational drug editing using retrieval and domain feedback. In 12th International Conference on Learning Representations (2024)

    2024

  6. [14]

    & Schmitt, S

    Bostr¨ om, J., Hogner, A. & Schmitt, S. Do structurally similar ligands bind in a similar fashion? J. Med. Chem.49, 6716–6725 (2006)

    2006

  7. [15]

    Eberhardt, J., Santos-Martins, D., Tillack, A. F. & Forli, S. Autodock vina 1.2.0: New docking methods, expanded force field, and python bindings. J. Chem. Inf. Model.61, 3891–3898 (2021)

    2021

  8. [16]

    R., Paolini, G

    Bickerton, G. R., Paolini, G. V., Besnard, J., Muresan, S. & Hopkins, A. L. Quantifying the chemical beauty of drugs. Nat. Chem. 4, 90–98 (2012)

    2012

  9. [17]

    A., Lombardo, F., Dominy, B

    Lipinski, C. A., Lombardo, F., Dominy, B. W. & Feeney, P. J. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv. Drug Deliv. Rev.23, 3–25 (1997)

    1997

  10. [18]

    Automatic chemical design using a data-driven continuous representation of molecules

    G´ omez-Bombarelli, R.et al. Automatic chemical design using a data-driven continuous representation of molecules. ACS Cent. Sci. 4, 268–276 (2018)

    2018

  11. [19]

    R., Parthasarathy, S

    Chen, Z., Min, M. R., Parthasarathy, S. & Ning, X. A deep generative model for molecule optimization via one fragment modification. Nat. Mach. Intell.3, 1040–1049 (2021)

    2021

  12. [20]

    G., Vignac, C

    Hoogeboom, E., Satorras, V. G., Vignac, C. & Welling, M. Equivariant diffusion for molecule generation in 3D. In Chaudhuri, K. et al. (eds.) Proceedings of the 39th International Conference on Machine Learning, vol. 162 of Proceedings of Machine Learning Research, 8867–8887 (P...

    2022

  13. [21]

    & Zhou, H

    Long, S., Zhou, Y., Dai, X. & Zhou, H. Zero-shot 3d drug design by sketching and generating. In Oh, A. H., Agarwal, A., Belgrave, D. & Cho, K. (eds.) Advances in Neural Information Processing Systems(2022)

    2022

  14. [22]

    & Coley, C

    Adams, K. & Coley, C. W. Equivariant shape-conditioned generation of 3d molecules for ligand-based drug design. In 11th International Conference on Learning Representations(2023)

    2023

  15. [23]

    & Ning, X

    Chen, Z., Peng, B., Parthasarathy, S. & Ning, X. Shape-conditioned 3d molecule generation via equivariant diffusion models. arXiv:2403.12987 (2023)

    2023 arXiv

  16. [24]

    & Noe, F

    K¨ ohler, J., Klein, L. & Noe, F. Equivariant flows: Exact likelihood generative learning for symmetric densities. In III, H. D. & Singh, A. (eds.) Proceedings of the 37th International Conference on Machine Learning, vol. 119 of Proceedings of Machine Learning Research, 5361–...

    2020

  17. [25]

    & Peng, J

    Luo, S., Guan, J., Ma, J. & Peng, J. A 3d generative model for structure-based drug design. In Beygelzimer, A., Dauphin, Y., Liang, P. & Vaughan, J. W. (eds.) Advances in Neural Information Processing Systems(2021)

    2021

  18. [26]

    et al.Pocket2Mol: Efficient molecular sampling based on 3D protein pockets

    Peng, X. et al.Pocket2Mol: Efficient molecular sampling based on 3D protein pockets. In Chaudhuri, K.et al.(eds.) Proceedings of the 39th International Conference on Machine Learning, vol. 162 of Proceedings of Machine Learning Research, 17644–17655 (PMLR, 2022)

    2022

  19. [27]

    Guan, J. et al. 3d equivariant diffusion for target-aware molecule generation and affinity prediction. In 11th International Conference on Learning Representations(2023)

    2023

  20. [28]

    et al.DecompDiff: Diffusion models with decomposed priors for structure-based drug design

    Guan, J. et al.DecompDiff: Diffusion models with decomposed priors for structure-based drug design. In Krause, A. et al.(eds.) Proceedings of the 40th International Conference on Machine Learning, vol. 202 of Proceedings of Machine Learning Research, 11827–11846 (PMLR, 2023)

    2023

  21. [29]

    & Koes, D

    Ragoza, M., Masuda, T. & Koes, D. R. Generating 3D molecules conditional on receptor binding sites with deep generative models. Chem. Sci. 13, 2701–2713 (2022)

    2022

  22. [30]

    Liu, M., Luo, Y., Uchino, K., Maruhashi, K. & Ji, S. Generating 3D molecules for target protein binding. In Chaudhuri, K. et al.(eds.) Proceedings of the 39th International Conference on Machine Learning, vol. 162 of Proceedings of Machine Learning Research, 13912–13924 (PMLR, 2022)

    2022

  23. [31]

    Polykovskiy, D. et al. Molecular sets (moses): A benchmarking platform for molecular generation models. Front. Pharmacol. 11 (2020)

    2020

  24. [32]

    Landrum, G. et al. rdkit/rdkit: 2023 03 2 (q1 2023) release (2023)

    2023

  25. [33]

    Francoeur, P. G. et al. Three-dimensional convolutional neural networks and a cross-docked data set for structure-based drug design. J. Chem. Inf. Model.60, 4200–4215 (2020)

    2020

  26. [34]

    Hawkins, P. C. D., Skillman, A. G. & Nicholls, A. Comparison of shape-matching and docking as virtual screening tools. J. Med. Chem. 50, 74–82 (2006)

    2006

  27. [35]

    & Schuffenhauer, A

    Ertl, P. & Schuffenhauer, A. Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions. J. Cheminform. 1, 8 (2009). 20

    2009

  28. [36]

    Peng, X., Guan, J., Liu, Q. & Ma, J. MolDiff: Addressing the atom-bond inconsistency problem in 3D molecule diffusion generation. In Krause, A. et al. (eds.) Proceedings of the 40th International Conference on Machine Learning, vol. 202 of Proceedings of Machine Learning Resea...

    2023

  29. [37]

    Tingle, B. I. et al. Zinc-22-a free multi-billion-scale database of tangible compounds for ligand discovery. J. Chem. Inf. Model. 63, 1166–1176 (2023)

    2023

  30. [38]

    & Andricopulo, A

    Ferreira, L., dos Santos, R., Oliva, G. & Andricopulo, A. Molecular docking and structure-based drug design strategies. Molecules 20, 13384–13421 (2015)

    2015

  31. [39]

    Tadesse, S., Yu, M., Kumarasiri, M., Le, B. T. & Wang, S. Targeting cdk6 in cancer: State of the art and new insights. Cell Cycle 14, 3220–3230 (2015)

    2015

  32. [40]

    El-Amouri, S. S. et al. Neprilysin: An enzyme candidate to slow the progression of alzheimer’s disease. Am. J. Pathol.172, 1342–1354 (2008)

    2008

  33. [41]

    Burley, S. K. et al. Rcsb protein data bank (rcsb.org): delivery of experimentally-determined pdb structures alongside one million computed structure models of proteins from artificial intelligence/machine learning. Nucleic Acids Res.51, D488–D508 (2022)

    2022

  34. [42]

    Neves, M. A. C., Totrov, M. & Abagyan, R. Docking and scoring with icm: the benchmarking results and strategies for improvement. J. Comput. Aided Mol. Des.26, 675–686 (2012)

    2012

  35. [43]

    Yang, H. et al. admetsar 2.0: web-service for prediction and optimization of chemical admet properties. Bioinform. 35, 1067–1069 (2018)

    2018

  36. [44]

    Halgren, T. A. Merck molecular force field. i. basis, form, scope, parameterization, and performance of mmff94. J. Comput. Chem. 17, 490–519 (1996)

    1996

  37. [45]

    Patnaik, A. et al. Efficacy and safety of abemaciclib, an inhibitor of cdk4 and cdk6, for patients with breast cancer, non–small cell lung cancer, and other solid tumors. Cancer Discov.6, 740–753 (2016)

    2016

  38. [46]

    Palbociclib: a first-in-class cdk4/cdk6 inhibitor for the treatment of hormone-receptor positive advanced breast cancer

    Lu, J. Palbociclib: a first-in-class cdk4/cdk6 inhibitor for the treatment of hormone-receptor positive advanced breast cancer. J. Hematol. Oncol.8, 98 (2015)

    2015

  39. [47]

    & Sellers, W

    Tripathy, D., Bardia, A. & Sellers, W. R. Ribociclib (lee011): Mechanism of action and clinical impact of this selective cyclin-dependent kinase 4/6 inhibitor in various solid tumors. Clin. Cancer Res.23, 3251–3262 (2017)

    2017

  40. [48]

    & Giuliani, A

    Benigni, R., Bossa, C., Tcheremenskaia, O. & Giuliani, A. Alternatives to the carcinogenicity bioassay:in silicomethods, and thein vitroandin vivomutagenicity assays. Expert Opin. Drug Metab. Toxicol.6, 809–819 (2010)

    2010

  41. [49]

    Y.-C., Jansen, J., Masereeuw, R

    Soo, J. Y.-C., Jansen, J., Masereeuw, R. & Little, M. H. Advances in predictive in vitro models of drug-induced nephrotoxicity. Nat. Rev. Nephrol.14, 378–393 (2018)

    2018

  42. [50]

    Hansen, R. A. et al. Efficacy and safety of donepezil, galantamine, and rivastigmine for the treatment of alzheimer’s disease: a systematic review and meta-analysis. Clin. Interv. Aging3, 211–225 (2008)

    2008

  43. [51]

    & Zlokovic, B

    Deane, R. & Zlokovic, B. Role of the blood-brain barrier in the pathogenesis of alzheimers disease. Curr. Alzheimer Res.4, 191–197 (2007)

    2007

  44. [52]

    Du, X. et al. Insights into protein–ligand interactions: Mechanisms, models, and methods. Int. J. Mol. Sci.17, 144 (2016)

    2016

  45. [53]

    Lin, J. et al. The role of absorption, distribution, metabolism, excretion and toxicity in drug discovery. Curr. Top. Med. Chem. 3, 1125–1154 (2003)

    2003

  46. [54]

    R., Fuchs, J

    Chen, Z., Ayinde, O. R., Fuchs, J. R., Sun, H. & Ning, X. G2retro as a two-step graph generative models for retrosynthesis prediction. Commun. Chem. 6, 102 (2023)

    2023

  47. [55]

    & Coley, C

    Gao, W. & Coley, C. W. The synthesizability of molecules proposed by generative models. J. Chem. Inf. Model.60, 5714–5723 (2020)

    2020

  48. [56]

    & Schneider, G

    Atz, K., Grisoni, F. & Schneider, G. Geometric deep learning on molecular representations. Nat. Mach. Intell.3, 1023–1032 (2021)

    2021

  49. [57]

    J., Florence, P., Straub, J., Newcombe, R

    Park, J. J., Florence, P., Straub, J., Newcombe, R. & Lovegrove, S. Deepsdf: Learning continuous signed distance functions for shape representation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2019)

    2019

  50. [58]

    Deng, C. et al. Vector neurons: A general framework for so(3)-equivariant networks. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 12200–12209 (2021)

    2021

  51. [59]

    Wang, Y. et al. Dynamic graph cnn for learning on point clouds. ACM Trans. Graph.38, 1–12 (2019)

    2019

  52. [60]

    & Abbeel, P

    Ho, J., Jain, A. & Abbeel, P. Denoising diffusion probabilistic models. In Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M. & Lin, H. (eds.) Advances in Neural Information Processing Systems, vol. 33, 6840–6851 (Curran Associates, Inc., 2020)

    2020

  53. [61]

    & Welling, M

    Hoogeboom, E., Nielsen, D., Jaini, P., Forr´ e, P. & Welling, M. Argmax flows and multinomial diffusion: Learning categorical distributions. In Ranzato, M., Beygelzimer, A., Dauphin, Y., Liang, P. & Vaughan, J. W. (eds.) Advances in Neural Information Processing Systems, vol. ...

    2021

  54. [62]

    & Leibler, R

    Kullback, S. & Leibler, R. A. On information and sufficiency. The Annals of Mathematical Statistics22, 79–86 (1951)

    1951

  55. [63]

    Jumper, J. et al. Highly accurate protein structure prediction with alphafold. Nat. 596, 583–589 (2021)

    2021

  56. [64]

    Jing, B., Eismann, S., Suriana, P., Townshend, R. J. L. & Dror, R. Learning from protein structure with geometric vector perceptrons. In 11th International Conference on Learning Representations(2021)

    2021

  57. [65]

    & Welling, M

    Garcia Satorras, V., Hoogeboom, E., Fuchs, F., Posner, I. & Welling, M. E(n) equivariant normalizing flows. In Ranzato, M., Beygelzimer, A., Dauphin, Y., Liang, P. & Vaughan, J. W. (eds.) Advances in Neural Information Processing Systems, vol. 34, 4181–4192 (Curran Associates,...

    2021

  58. [66]

    Torge, J., Harris, C., Mathis, S. V. & Lio, P. Diffhopp: A graph diffusion model for novel drug design via scaffold hopping. arXiv:2308.07416 (2023)

    2023 arXiv

  59. [67]

    & Nichol, A

    Dhariwal, P. & Nichol, A. Q. Diffusion models beat GANs on image synthesis. In Beygelzimer, A., Dauphin, Y., Liang, P. & Vaughan, J. W. (eds.) Advances in Neural Information Processing Systems(2021)

    2021

  60. [68]

    Kingma, D. P. & Ba, J. Adam: A method for stochastic optimization. In Bengio, Y. & LeCun, Y. (eds.) 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, 2015(2015)

    2015

  61. [69]

    Portal, O. C. logs calculation. URL https://www.organic-chemistry.org/prog/peo/logS.html

  62. [70]

    & Siedlecki, P

    W´ ojcikowski, M., Zielenkiewicz, P. & Siedlecki, P. Open drug discovery toolkit (ODDT): a new open-source player in the drug discovery field. J. Cheminform. 7 (2015). 21

    2015

  63. [71]

    Ravi, N. et al. Accelerating 3d deep learning with pytorch3d. arXiv:2007.08501 (2020)

    2020 arXiv

  64. [72]

    Nichol, A. Q. & Dhariwal, P. Improved denoising diffusion probabilistic models. In Meila, M. & Zhang, T. (eds.) Proceedings of the 38th International Conference on Machine Learning, vol. 139 of Proceedings of Machine Learning Research, 8162–8171 (PMLR, 2021)

    2021

  65. [73]

    H”, “C”, “N

    Kong, Z., Ping, W., Huang, J., Zhao, K. & Catanzaro, B. Diffwave: A versatile diffusion model for audio synthesis. In International Conference on Learning Representations(2021). 22 Generating 3D Binding Molecules Using Shape-Conditioned Diffusion Models with Guidance (Suppleme...

    2021

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.