CoChat
Sign Up
Cited by 7Enzyme Structure and FunctionComputational Drug Discovery MethodsAdvanced Electron Microscopy Techniques and Applications

The DOI matches its Crossref record; 3 of the 3 references it cites were found in a registry or an index.

Ligand identification in CryoEM and X-ray maps using deep learning

Open access
Karolczak, Jacek, Przybyłowska, Anna, Szewczyk, Konrad, Taisner, Witold, Heumann, John M, Stowell, Michael H B, Nowicki, Michał, Brzezinski, Dariusz·2024·Bioinformatics
DOI 10.1093/bioinformatics/btae749PMC PMC11709248

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

Motivation: Accurately identifying ligands plays a crucial role in the process of structure-guided drug design. Based on density maps from X-ray diffraction or cryogenic-sample electron microscopy (cryoEM), scientists verify whether small-molecule ligands bind to active sites of interest. However, the interpretation of density maps is challenging, and cognitive bias can sometimes mislead investigators into modeling fictitious compounds. Ligand identification can be aided by automatic methods, but existing approaches are available only for X-ray diffraction and are based on iterative fitting or feature-engineered machine learning rather than end-to-end deep learning. Results: Here, we propose to identify ligands using a deep-learning approach that treats density maps as 3D point clouds. We show that the proposed model is on par with existing machine learning methods for X-ray crystallography while also being applicable to cryoEM density maps. Our study demonstrates that electron density map fragments can aid the training of models that can later be applied to cryoEM structures but also highlights challenges associated with the standardization of electron microscopy maps and the quality assessment of cryoEM ligands. Availability and implementation: Code and model weights are available on GitHub at https://github.com/jkarolczak/ligands-classification. An accompanying ChimeraX bundle is available at https://github.com/wtaisner/chimerax-ligand-recognizer.

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 · 10.1093/bioinformatics/btae749

Cites

3 of 3 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1. Illustration of the density thresholding used for cryoEM maps. (A) The default voxel value distribution for PDB deposit 7SMR. The distribution is zero-inflated (almost 15 million voxels with values close to zero). (B) Voxel value distribution of 7SMR after the removal of values ±0.5 standard deviation around the median. The resulting distribution is much closer to normal and the quantile threshold is over two times higher.
Figure 1. Illustration of the density thresholding used for cryoEM maps. (A) The default voxel value distribution for PDB deposit 7SMR. The distribution is zero-inflated (almost 15 million voxels with values close to zero). (B) Voxel value distribution of 7SMR after the removal of values ±0.5 standard deviation around the median. The resulting distribution is much closer to normal and the quantile threshold is over two times higher.CC BY · PubMed Central
Figure 2. Schematic representation of the proposed ligand recognition pipeline. Example presenting the processing of a nicotinamide adenine dinucleotide ligand from 2.10 Å cryoEM PDB deposit 8fuz (residue A 602). An initially selected fragment of a difference map (represented here at a 2.8σ threshold) is given a new density threshold based on reducing zero inflation and quantile thresholding. The map fragment is then resampled into a 0.2 Å grid, voxel values are rescaled, and the grid is transformed into a point cloud. From the point cloud, 2000 points are selected (using uniform maximum point selection) and fed into a deep learning model that outputs a list of ligand group scores.
Figure 2. Schematic representation of the proposed ligand recognition pipeline. Example presenting the processing of a nicotinamide adenine dinucleotide ligand from 2.10 Å cryoEM PDB deposit 8fuz (residue A 602). An initially selected fragment of a difference map (represented here at a 2.8σ threshold) is given a new density threshold based on reducing zero inflation and quantile thresholding. The map fragment is then resampled into a 0.2 Å grid, voxel values are rescaled, and the grid is transformed into a point cloud. From the point cloud, 2000 points are selected (using uniform maximum point selection) and fed into a deep learning model that outputs a list of ligand group scores.CC BY · PubMed Central
Figure 3. Predictive performance of the analyzed machine learning models for X-ray (top panels), cryoEM (middle panels), and mixed (bottom panels) test sets of ligands. Each plot shows the mean ranks of the correct ligand (with error bars showing the standard error) versus (A, D, G) map resolution, (B, E, H) ligand fit quality (RSCC for X-ray, Q-score for cryoEM), and (C, F, I) size (number of nonhydrogen atoms) of the ligand. The corresponding ligand distributions are depicted as gray histograms, with ligand counts shown on the right y-axis.
Figure 3. Predictive performance of the analyzed machine learning models for X-ray (top panels), cryoEM (middle panels), and mixed (bottom panels) test sets of ligands. Each plot shows the mean ranks of the correct ligand (with error bars showing the standard error) versus (A, D, G) map resolution, (B, E, H) ligand fit quality (RSCC for X-ray, Q-score for cryoEM), and (C, F, I) size (number of nonhydrogen atoms) of the ligand. The corresponding ligand distributions are depicted as gray histograms, with ligand counts shown on the right y-axis.CC BY · PubMed Central
Figure 4. Examples of ligand identification using the proposed MinkLoc3Dv2 model. (A–D) Examples of correctly predicted X-ray ligands. (E) Uridine-5ʹ-diphosphate (UDP) misclassified as uridine (URI, black dashed frame). (F–I) Examples of correctly predicted cryoEM ligands. (J) Heme A (HEM) misclassified as a rare ligand due to incorrect density thresholding. Each ligand is labeled by its Chemical Component Dictionary ID, structure resolution, and (in parentheses) the PDB ID, chain, and residue number. X-ray diffraction ligands (A–E) shown in green mesh based on Fo–Fc maps contoured at 2.8σ calculated after removal of solvent and other small molecules (including the ligand) from the model. CryoEM ligands (F–J) depicted in pink mesh based on difference maps contoured according to the proposed automatic density thresholding method (13.642, 3.385, 17.997, 7.850, and 5.613 V for panels F–J, respectively). The white mesh in panel J shows a manually selected contour threshold of 11.000 V. Atomic coordinates were taken from the PDB deposits.
Figure 4. Examples of ligand identification using the proposed MinkLoc3Dv2 model. (A–D) Examples of correctly predicted X-ray ligands. (E) Uridine-5ʹ-diphosphate (UDP) misclassified as uridine (URI, black dashed frame). (F–I) Examples of correctly predicted cryoEM ligands. (J) Heme A (HEM) misclassified as a rare ligand due to incorrect density thresholding. Each ligand is labeled by its Chemical Component Dictionary ID, structure resolution, and (in parentheses) the PDB ID, chain, and residue number. X-ray diffraction ligands (A–E) shown in green mesh based on Fo–Fc maps contoured at 2.8σ calculated after removal of solvent and other small molecules (including the ligand) from the model. CryoEM ligands (F–J) depicted in pink mesh based on difference maps contoured according to the proposed automatic density thresholding method (13.642, 3.385, 17.997, 7.850, and 5.613 V for panels F–J, respectively). The white mesh in panel J shows a manually selected contour threshold of 11.000 V. Atomic coordinates were taken from the PDB deposits.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

Show all 3 references

Research works printed but not linked: 42

  1. Afonine, 2018, New tools for the analysis and validation of cryo-EM maps and atomic models, Acta Crystallogr D Struct Biol, 74, 814, 10.1107/S2059798318009324 · DOI
  2. Aishima, 2005, Automated crystallographic ligand building using the medial axis transform of an electron-density isosurface, Acta Crystallogr D Biol Crystallogr, 61, 1354, 10.1107/S0907444905023152 · DOI
  3. Banko, 2021, Deep learning for visualization and novelty detection in large X-ray diffraction datasets, NPJ Comput Mater, 7, 104, 10.1038/s41524-021-00575-9 · DOI
  4. Beckers, 2019, Thresholding of cryo-EM density maps by false discovery rate control, IUCrJ, 6, 18, 10.1107/S2052252518014434 · DOI
  5. Bepler, 2019, Positive-unlabeled convolutional neural networks for particle picking in cryo-electron micrographs, Nat Methods, 16, 1153, 10.1038/s41592-019-0575-8 · DOI
  6. Berman, 2000, The protein data bank, Nucleic Acids Res, 28, 235, 10.1093/nar/28.1.235 · DOI · Paper page
  7. Brzezinski, 2021, Recognizing and validating ligands with CheckMyBlob, Nucleic Acids Res, 49, W86, 10.1093/nar/gkab296 · DOI
  8. Burley, 2022, Electron microscopy holdings of the protein data bank: the impact of the resolution revolution, new validation tools, and implications for the future, Biophys Rev, 14, 1281, 10.1007/s12551-022-01013-w · DOI
  9. Carolan, 2014, Automated identification of crystallographic ligands using sparse-density representations, Acta Crystallogr D Biol Crystallogr, 70, 1844, 10.1107/S1399004714008578 · DOI
  10. Czyzewski, 2021, Detecting anomalies in X-ray diffraction images using convolutional neural networks, Expert Syst Appl, 174, 114740, 10.1016/j.eswa.2021.114740 · DOI
  11. Deller, 2015, Models of protein–ligand crystal structures: trust, but verify, J Comput Aided Mol Des, 29, 817, 10.1007/s10822-015-9833-8 · DOI
  12. Dhakal, 2024, CryoTransformer: a transformer model for picking protein particles from Cryo-EM micrographs, Bioinformatics, 40, 10.1093/bioinformatics/btae109 · DOI
  13. Evrard, 2007, Assessment of automatic ligand building in ARP/wARP, Acta Crystallogr D Biol Crystallogr, 63, 108, 10.1107/S0907444906023389 · DOI
  14. Gunasekaran, 2009, Pattern Recognition in Bioinformatics, Lecture Notes in Computer Science, 125, 10.1007/978-3-642-04031-3_12 · DOI
  15. He, 2023, Improvement of cryo-EM maps by simultaneous local and non-local deep learning, Nat Commun, 14, 3217, 10.1038/s41467-023-39031-1 · DOI
  16. Ito, 2019, DeepCentering: fully automated crystal centering using deep learning for macromolecular crystallography, J Synchrotron Radiat, 26, 1361, 10.1107/S160057751900434X · DOI
  17. Japkowicz, 2011, Evaluating Learning Algorithms: A Classification Perspective, 10.1017/CBO9780511921803 · DOI · Paper page
  18. Kowiel, 2019, Automatic recognition of ligands in electron density by machine learning, Bioinformatics, 35, 452, 10.1093/bioinformatics/bty626 · DOI
  19. Lawson, 2024, Outcomes of the EMDataResource cryo-EM ligand modeling challenge, Nat Methods, 21, 1340, 10.1038/s41592-024-02321-7 · DOI
  20. Li, 2024, All-atom RNA structure determination from cryo-EM maps, Nat Biotechnol, 10.1038/s41587-024-02149-8 · DOI
  21. Liebschner, 2019, Macromolecular structure determination using X-rays, neutrons and electrons: recent developments in phenix, Acta Crystallogr D Struct Biol, 75, 861, 10.1107/S2059798319011471 · DOI · Paper page
  22. Marques, 2019, CryoEM maps are full of potential, Curr Opin Struct Biol, 58, 214, 10.1016/j.sbi.2019.04.006 · DOI
  23. Matinyan, 2024, Deep learning applications in protein crystallography, Acta Crystallogr A Found Adv, 80, 1, 10.1107/S2053273323009300 · DOI
  24. Misiura, 2022, DLPacker : deep learning for prediction of amino acid side chain conformations in proteins, Proteins, 90, 1278, 10.1002/prot.26311 · DOI · Paper page
  25. Mitsuoka, 1999, The structure of bacteriorhodopsin at 3.0 Å resolution based on electron crystallography: implication of the charge distribution 1 1Edited by R. Huber, J Mol Biol, 286, 861, 10.1006/jmbi.1998.2529 · DOI
  26. Muenks, 2023, Automatic and accurate ligand structure determination guided by cryo-electron microscopy maps, Nat Commun, 14, 1164, 10.1038/s41467-023-36732-5 · DOI · Paper page
  27. Pan, 2023, A deep learning solution for crystallographic structure determination, IUCrJ, 10, 487, 10.1107/S2052252523004293 · DOI
  28. Pearce, 2017, A multi-crystal method for extracting obscured crystallographic states from conventionally uninterpretable electron density, Nat Commun, 8, 15123, 10.1038/ncomms15123 · DOI
  29. Pintilie, 2020, Measurement of atom resolvability in cryo-EM maps with Q-scores, Nat Methods, 17, 328, 10.1038/s41592-020-0731-1 · DOI
  30. Rosenthal, 2003, Optimal determination of particle orientation, absolute hand, and contrast loss in single-particle electron cryomicroscopy, J Mol Biol, 333, 721, 10.1016/j.jmb.2003.07.013 · DOI · Paper page
  31. Sanchez-Garcia, 2021, DeepEMhancer: a deep learning solution for cryo-EM volume post-processing, Commun Biol, 4, 874, 10.1038/s42003-021-02399-1 · DOI
  32. Smart, 2018, Validation of ligands in macromolecular structures determined by X-ray crystallography, Acta Crystallogr D Struct Biol, 74, 228, 10.1107/S2059798318002541 · DOI
  33. Terwilliger, 2006, Automated ligand fitting by core-fragment fitting and extension into density, Acta Crystallogr D Biol Crystallogr, 62, 915, 10.1107/S0907444906017161 · DOI
  34. Terwilliger, 2007, Ligand identification using electron-density map correlations, Acta Crystallogr D Biol Crystallogr, 63, 101, 10.1107/S0907444906046233 · DOI
  35. Turner, 2024, EMDB—the electron microscopy data bank, Nucleic Acids Res, 52, D456, 10.1093/nar/gkad1019 · DOI
  36. Wang, 2017, Experimental charge density from electron microscopic maps, Protein Sci, 26, 1619, 10.1002/pro.3198 · DOI
  37. Weiler, 2018, 32nd Conference on Neural Information Processing Systems (NeurIPS 2018). Montréal, Canada
  38. Wlodawer, 2018, Detect, correct, retract: how to manage incorrect structural models, Febs J, 285, 444, 10.1111/febs.14320 · DOI
  39. Xu, 2023, TransLoc3D: point cloud based large-scale place recognition using adaptive receptive fields, Commun Inf Syst, 23, 57, 10.4310/CIS.2023.v23.n1.a3 · DOI
  40. Zhang, 2022, RIConv++: effective rotation invariant convolutions for 3D point clouds deep learning, Int J Comput Vis, 130, 1228, 10.1007/s11263-022-01601-z · DOI
  41. Zwart, 2004, Modelling bound ligands in protein crystal structures, Acta Crystallogr D Biol Crystallogr, 60, 2230, 10.1107/S0907444904012995 · DOI
  42. Zywanowski, 2022, MinkLoc3D-SI: 3D LiDAR place recognition with sparse convolutions, spherical coordinates, and intensity, IEEE Robot Autom Lett, 7, 1079, 10.1109/LRA.2021.3136863 · DOI

Metadata

Authors
Karolczak, J., Przybyłowska, A., Szewczyk, K., Taisner, W., Heumann, J. M., Stowell, M. H. B., Nowicki, M., & Brzezinski, D.
Affiliations
  • Poznań University of Technology (PL)
  • University of Colorado Boulder (US)
  • Institute of Computing Science, Poznan University of Technology , Poznan 60-965,
ORCID
Venue
Bioinformatics
Publisher
Oxford University Press (OUP)
Published
2024-12-19
Volume
41
Issue
1
ISSN
  • 1367-4803 (ISSN-L)
  • 1367-4811
DOI
10.1093/bioinformatics/btae749
PMC
PMC11709248
OpenAlex
W4405596261
Cited by
7
References
3
Citations by year
2026: 3 · 2025: 4
FWCI
0.68
Type
journal article
Language
en
Open access
Licence
Locations
  • Bioinformatics (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (submittedVersion, cc-by)
Funders
NIH (AG061829, NS120496, U24-GM139174)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Structural Biology › Enzyme Structure and Function
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Structural Biology › Advanced Electron Microscopy Techniques and Applications
Keywords
ligand identification, 3D point clouds, deep learning, ligand classification, structure-based drug design, map standardization

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Ligand identification in CryoEM and X-ray maps using deep learning. https://app.cochat.ai/papers/10.1093/bioinformatics/btae749
The paper
Karolczak, J., Przybyłowska, A., Szewczyk, K., Taisner, W., Heumann, J. M., Stowell, M. H. B., Nowicki, M., & Brzezinski, D. (2024). Ligand identification in CryoEM and X-ray maps using deep learning. Bioinformatics. https://doi.org/10.1093/bioinformatics/btae749