CoChat
Sign Up
Cited by 21Computational Drug Discovery MethodsProtein Degradation and InhibitorsMachine Learning in Materials Science

The DOI matches its Crossref record; 10 of the 10 references it cites were found in a registry or an index.

DiffInt: A Diffusion Model for Structure-Based Drug Design with Explicit Hydrogen Bond Interaction Guidance

Open access
Sako, Masami, Yasuo, Nobuaki, Sekijima, Masakazu·2025·Journal of Chemical Information and Modeling
DOI 10.1021/acs.jcim.4c01385PMC PMC11733934

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

The design of drug molecules is a critical stage in the drug discovery process. The structure-based drug design has long played an important role in efficient development. Significant progress has been made in recent years in the generation of 3D molecules via deep generation models. However, while many existing models have succeeded in incorporating structural information on target proteins, they have not been able to address important interactions between protein and drug molecules, especially hydrogen bonds. In this study, we propose DiffInt as a novel structure-based approach that explicitly addresses interactions. The model naturally incorporates hydrogen bonds between protein and ligand molecules by treating them as pseudoparticles. The experimental results show that DiffInt reproduces hydrogen bonds, and the hydrogen binding energies significantly outperform those of existing models. To facilitate the use of our tool for generating new drug molecules based on any protein's three-dimensional structure, we have made the source code and trained model available on GitHub (https://github.com/sekijima-lab/DiffInt) under the MIT license, with the execution environment provided on Google Colab.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1021/acs.jcim.4c01385

Cites

10 of 10 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Overview of DiffInt. The diffusion process adds noise to the data to obtain a latent noise representation, resulting in a Gaussian distribution at time step T in the center-of-gravity system of the input data. The generative process reconstructs the original data by gradually denoising the noise, and consequently also reconstructs the hydrogen bonds with the protein pocket. In both processes, the protein molecules and hydrogen bonding particles are treated as conditions and remain unchanged.
Figure 1 Overview of DiffInt. The diffusion process adds noise to the data to obtain a latent noise representation, resulting in a Gaussian distribution at time step T in the center-of-gravity system of the input data. The generative process reconstructs the original data by gradually denoising the noise, and consequently also reconstructs the hydrogen bonds with the protein pocket. In both processes, the protein molecules and hydrogen bonding particles are treated as conditions and remain unchanged.CC BY · PubMed Central
Figure 2 Introduction of interaction particles by considering hydrogen bonds as pseudoparticles. Two interaction particles are placed on each hydrogen bond (yellow dashed line) between the protein and ligand to provide information about the position and direction of the hydrogen bond. The red and yellow spheres indicate whether the ligand atom is electron-acceptor or electron-donor, respectively.
Figure 2 Introduction of interaction particles by considering hydrogen bonds as pseudoparticles. Two interaction particles are placed on each hydrogen bond (yellow dashed line) between the protein and ligand to provide information about the position and direction of the hydrogen bond. The red and yellow spheres indicate whether the ligand atom is electron-acceptor or electron-donor, respectively.CC BY · PubMed Central
Figure 3 Visualization of ligands with the target proteins (PDB IDs: 3cqw and 3c4f). (a,d) The reference ligand molecules binding to the proteins 3cqw and 4c4f, respectively. (b,c) Example ligands generated by the DiffInt for 3cqw. (e,f) Example ligands generated by the DiffInt for 3c4f. Upper panel in each figure shows 3D structures of the protein–ligand complexes and lower panel illustrates protein–ligand interaction.
Figure 3 Visualization of ligands with the target proteins (PDB IDs: 3cqw and 3c4f). (a,d) The reference ligand molecules binding to the proteins 3cqw and 4c4f, respectively. (b,c) Example ligands generated by the DiffInt for 3cqw. (e,f) Example ligands generated by the DiffInt for 3c4f. Upper panel in each figure shows 3D structures of the protein–ligand complexes and lower panel illustrates protein–ligand interaction.CC BY · PubMed Central
Figure 4 Prediction accuracy of hydrogen bond reproducibility for the four models. Each data point represents the average of hydrogen bond reproducibility for 100 molecules generated per pocket.
Figure 4 Prediction accuracy of hydrogen bond reproducibility for the four models. Each data point represents the average of hydrogen bond reproducibility for 100 molecules generated per pocket.CC BY · PubMed Central
Figure 5 Energy distribution of the best model of high affinity, displayed for each of the four models. Proportion of best models: Pocket2Mol: 4%, FLAG: 12%, DiffSBDD: 27%, DiffInt: 57%.
Figure 5 Energy distribution of the best model of high affinity, displayed for each of the four models. Proportion of best models: Pocket2Mol: 4%, FLAG: 12%, DiffSBDD: 27%, DiffInt: 57%.CC BY · PubMed Central
Figure 6 Number of data points becoming the best model for each model across strong, moderate, weak, and total regions.
Figure 6 Number of data points becoming the best model for each model across strong, moderate, weak, and total regions.CC BY · PubMed Central
Figure 7 Bond angle distributions for generated molecules and those in the test set. Density plots of the bond angles for (a) CCC, (b) CCO, (c) CNC, (d) OPO, (e) NCC, (f) CC=O, and (g) COC. Distributions are shown for molecules generated by four models (Pocket2Mol, FLAG, DiffSBDD, and DiffInt) and compared to the test set.
Figure 7 Bond angle distributions for generated molecules and those in the test set. Density plots of the bond angles for (a) CCC, (b) CCO, (c) CNC, (d) OPO, (e) NCC, (f) CC=O, and (g) COC. Distributions are shown for molecules generated by four models (Pocket2Mol, FLAG, DiffSBDD, and DiffInt) and compared to the test set.CC BY · PubMed Central
Figure 8 Dihedral angle distributions for generated molecules and those in the test set. Density plots of the dihedral angles for (a) CCCC, (b) cccc, (c) CCCO, (d) OCCO, (e) Cccc, and (f) CC=CC. Distributions are shown for molecules generated by four models (Pocket2Mol, FLAG, DiffSBDD, and DiffInt) and compared to the test set. The lower letters represent the atoms in the aromatic rings.
Figure 8 Dihedral angle distributions for generated molecules and those in the test set. Density plots of the dihedral angles for (a) CCCC, (b) cccc, (c) CCCO, (d) OCCO, (e) Cccc, and (f) CC=CC. Distributions are shown for molecules generated by four models (Pocket2Mol, FLAG, DiffSBDD, and DiffInt) and compared to the test set. The lower letters represent the atoms in the aromatic rings.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 75

  1. 10.1002/anie.198607673 · DOI
  2. 10.1038/326347a0 · DOI
  3. 10.1126/science.257.5073.1078 · DOI
  4. 10.1016/S0969-2126(00)00060-5 · DOI
  5. 10.1021/jm00090a001 · DOI
  6. 10.1007/BF00141577 · DOI
  7. 10.1021/ci950273r · DOI
  8. 10.1021/ci800064z · DOI
  9. 10.1021/ci800105p · DOI
  10. 10.1073/pnas.0604348103 · DOI
  11. 10.1016/S1359-6446(02)02483-2 · DOI
  12. 10.1038/nature03197 · DOI · Paper page
  13. 10.1093/nar/28.1.235 · DOI · Paper page
  14. 10.1021/jm030580l · DOI
  15. 10.1021/ci049714+ · DOI · Paper page
  16. 10.1093/nar/gkj067 · DOI · Paper page
  17. 10.1093/nar/gkp456 · DOI
  18. 10.1093/nar/gkr777 · DOI · Paper page
  19. 10.1038/181662a0 · DOI
  20. 10.1038/185416a0 · DOI
  21. 10.1017/S0033583500004297 · DOI
  22. 10.1126/science.abj8754 · DOI · Paper page
  23. 10.1038/s41592-022-01488-1 · DOI · Paper page
  24. 10.1007/s10822-013-9672-4 · DOI · Paper page
  25. 10.1021/ci00057a005 · DOI · Paper page
  26. 10.1088/2632-2153/aba947 · DOI
  27. 10.1021/acscentsci.7b00572 · DOI · Paper page
  28. Kusner M. J., 2017, International conference on machine learning, 1945
  29. Jin W., 2018, International conference on machine learning, 2323
  30. Guimaraes G. L., 2017, arXiv preprint arXiv:1705.10843 · arXiv
  31. De Cao N., 2018, arXiv preprint arXiv:1805.11973 · arXiv
  32. Madhawa K., 2019, arXiv preprint arXiv:1905.11600 · arXiv
  33. Shi C., 2020, arXiv preprint arXiv:2001.09382 · arXiv
  34. 10.1186/s13321-017-0235-x · DOI
  35. 10.1080/14686996.2017.1401424 · DOI
  36. 10.1186/s13321-021-00572-6 · DOI
  37. Honda S., 2019, arXiv preprint arXiv:1911.04738 · arXiv
  38. Dai H., 2018, arXiv preprint arXiv:1802.08786 · arXiv
  39. You J., 2018, Advances in neural information processing systems, 31
  40. Gebauer N., 2019, Advances in neural information processing systems, 32
  41. Simm G. N., 2019, arXiv preprint arXiv:1909.11459 · arXiv
  42. 10.1021/acs.jcim.1c00679 · DOI
  43. 10.1021/acsomega.3c05430 · DOI
  44. 10.1021/acs.jcim.0c01494 · DOI · Paper page
  45. 10.1021/acs.molpharmaceut.9b00634 · DOI
  46. 10.1039/D1SC05976A · DOI
  47. Peng X., 2022, International Conference on Machine Learning, 17644
  48. Zhang Z., 2023, Eleventh International Conference on Learning Representations
  49. 10.1039/D1SC04444C · DOI
  50. Hoogeboom E., 2022, International conference on machine learning, 8867
  51. Sohl-Dickstein J., 2015, International conference on machine learning, 2256
  52. Ho J., 2020, Advances in neural information processing systems, 33, 6840
  53. Schneuing A., 2022, arXiv preprint arXiv:2210.13695 · arXiv
  54. 10.1038/s41467-024-46569-1 · DOI
  55. 10.1021/acs.jcim.3c01466 · DOI · Paper page
  56. Guan J., 2023, Eleventh International Conference on Learning Representations
  57. 10.1038/s42256-024-00815-9 · DOI
  58. 10.1021/jm00145a002 · DOI
  59. 10.1002/(SICI)1096-987X(19981115)19:14<1639::AID-JCC10>3.0.CO;2-B · DOI
  60. 10.1006/jmbi.1999.3371 · DOI
  61. Kingma D., 2021, Advances in neural information processing systems, 34, 21696
  62. 10.1021/acs.jcim.0c00411 · DOI
  63. Luo S., 2021, Advances in Neural Information Processing Systems, 34, 6229
  64. 10.1038/nbt.3988 · DOI · Paper page
  65. 10.1186/s13321-015-0078-2 · DOI
  66. 10.1021/jm0306430 · DOI · Paper page
  67. 10.1021/jm051256o · DOI
  68. 10.1021/acs.jcim.8b00673 · DOI
  69. 10.1038/nchem.1243 · DOI
  70. 10.1186/1758-2946-1-8 · DOI · Paper page
  71. 10.1021/ci990307l · DOI
  72. 10.1021/cc9800071 · DOI
  73. 10.1016/S0169-409X(96)00423-1 · DOI
  74. 10.1186/s13321-015-0069-3 · DOI
  75. 10.1016/j.bspc.2021.102704 · DOI

Metadata

Authors
Sako, M., Yasuo, N., & Sekijima, M.
Affiliations
Institute of Science Tokyo (JP)
Corresponding author
Sekijima, Masakazu
ORCID
Venue
Journal of Chemical Information and Modeling
Publisher
American Chemical Society (ACS)
Published
2024-12-19
Volume
65
Issue
1
Pages
71-82
ISSN
  • 1549-9596 (ISSN-L)
  • 1549-960X
DOI
10.1021/acs.jcim.4c01385
PMC
PMC11733934
OpenAlex
W4405592636
Cited by
21
References
10
Citations by year
2026: 15 · 2025: 5 · 2024: 1
FWCI
4.45
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of Chemical Information and Modeling (publishedVersion, cc-by)
  • ChemRxiv (acceptedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
  • Japan Agency for Medical Research and Development (JP24ama121026)
  • Japan Society for the Promotion of Science (JP20H00620, JP24H01760)
Topics
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Degradation and Inhibitors
  • Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
Keywords
structure-based drug design, diffusion models, 3D molecule generation, hydrogen bonding, protein-ligand interactions, hydrogen binding energy, pseudoparticles
MeSH terms
Diffusion, Hydrogen Bonding, Ligands, Models, Molecular, Protein Conformation, Proteins, Software, Drug Design

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review