CoChat

Accurate prediction of transition metal ion location via deep learning

Open access
Dürr, Simon L., Levy, Andrea, Rothlisberger, Ursula · 2022 · Nature Communications

Preprint — published as 10.1038/s41467-023-37870-6 · Paper page

The DOI matches its Crossref record; 1 of the 1 references it cites were found in a registry or an index.

Abstract

Metal ions are essential cofactors for many proteins. In fact, currently, about half of the structurally characterized proteins contain a metal ion. Metal ions play a crucial role for many applications such as enzyme design or design of protein-protein interactions because they are biologically abundant, tether to the protein using strong interactions, and have favorable catalytic properties e.g. as Lewis acid. Computational design of metalloproteins is however hampered by the complex electronic structure of many biologically relevant metals such as zinc that can often not be accurately described using a classical force field. In this work, we develop two tools - Metal3D (based on 3D convolutional neural networks) and Metal1D (solely based on geometric criteria) to improve the identification and localization of zinc and other metal ions in experimental and computationally predicted protein structures. Comparison with other currently available tools shows that Metal3D is the most accurate metal ion location predictor to date outperforming geometric predictors including Metal1D by a wide margin using a single structure as input. Metal3D outputs a confidence metric for each predicted site and works on proteins with few homologes in the protein data bank. The predicted metal ion locations for Metal3D are within 0.70 ± 0.64 Å of the experimental locations with half of the sites below 0.5 Å. Metal3D predicts a global metal density that can be used for annotation of structures predicted using e.g. AlphaFold2 and a per residue metal density that can be used in protein design workflows for the location of suitable metal binding sites and rotamer sampling to create novel metalloproteins. Metal3D is available as easy to use webapp, notebook or commandline interface.

Overview

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

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

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

Integrity signals

Registry

Crossref · 10.1101/2022.08.22.504853

Cites

1 of 1 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Workflow of Metal3D and Metal1D. a Training of Metal3D and Metal1D is based on experimental Zn2+ sites. Metal1D extracts coordination environments from LINK records, Metal3D is a fully convolutional 3DCNN trained to predict the metal density from voxelized protein environments. b In inference mode Metal3D predicts the location of a metal ion by computing per residue metal densities and then averaging them to obtain a global metal density for the input proteins. The ions can then be placed using the weighted average of voxels above a cutoff. For Metal1D all residues in the protein are scanned for compatibility with the probability map. Metals are placed at the geometric center of residues with high scores according to the probability map. A final ranking of sites is obtained using the probability map.
Fig. 1 Workflow of Metal3D and Metal1D. a Training of Metal3D and Metal1D is based on experimental Zn2+ sites. Metal1D extracts coordination environments from LINK records, Metal3D is a fully convolutional 3DCNN trained to predict the metal density from voxelized protein environments. b In inference mode Metal3D predicts the location of a metal ion by computing per residue metal densities and then averaging them to obtain a global metal density for the input proteins. The ions can then be placed using the weighted average of voxels above a cutoff. For Metal1D all residues in the protein are scanned for compatibility with the probability map. Metals are placed at the geometric center of residues with high scores according to the probability map. A final ranking of sites is obtained using the probability map.CC BY · PubMed Central
Fig. 2 Metal3D probability density. Probability evolution in HCA2 (PDB 2CBA) for different probability cutoffs. Isosurfaces highlighted at different cutoffs (a) p = 0.1 (b) p = 0.2 (c) p = 0.9 (d) inset on active site of HCA2 p = 0.9 Predicted positions in darkpurple, experimental position in green, crystal water in red.
Fig. 2 Metal3D probability density. Probability evolution in HCA2 (PDB 2CBA) for different probability cutoffs. Isosurfaces highlighted at different cutoffs (a) p = 0.1 (b) p = 0.2 (c) p = 0.9 (d) inset on active site of HCA2 p = 0.9 Predicted positions in darkpurple, experimental position in green, crystal water in red.CC BY · PubMed Central
Fig. 3 Identification of metal sites within 5 Å. a Comparison of Metal1D, Metal3D, BioMetAll, MIB, MIB2 on the test set held out from training of Metal1D and Metal3D. Predicted sites are counted as true positives (TP) if they are within 5 Å of a true metal location and as false negatives (FN) otherwise. False positive (FP) probes are clustered and counted once per cluster. The main parameter for each method (t Metal1d, p Metal3D, c BioMetAll, t MIB/MIB2) is explained in Table S5 *For MIB/MIB2 we used 2 structures less because the server did not accept these structures. b Precision-Recall curve for all tested tools for the binding site task split into all zinc sites in the test set (dotted lines), sites with 2+ unique coordination partners (dashed lines) and sites with 3+ unique coordination partners (solid lines).
Fig. 3 Identification of metal sites within 5 Å. a Comparison of Metal1D, Metal3D, BioMetAll, MIB, MIB2 on the test set held out from training of Metal1D and Metal3D. Predicted sites are counted as true positives (TP) if they are within 5 Å of a true metal location and as false negatives (FN) otherwise. False positive (FP) probes are clustered and counted once per cluster. The main parameter for each method (t Metal1d, p Metal3D, c BioMetAll, t MIB/MIB2) is explained in Table S5 *For MIB/MIB2 we used 2 structures less because the server did not accept these structures. b Precision-Recall curve for all tested tools for the binding site task split into all zinc sites in the test set (dotted lines), sites with 2+ unique coordination partners (dashed lines) and sites with 3+ unique coordination partners (solid lines).CC BY · PubMed Central
Fig. 4 Mean absolute deviation of correctly predicted sites and selectivity for other ions. a For all sites where predicted and experimental location are available (true positives in Fig. 3) we compute the mean absolute deviation (MAD) using Metal1D, Metal3D, BioMetAll, MIB and MIB2 on the test set. Some tools predict multiple locations (n) per site (TP). *For MIB and MIB2 we used 2 structures less because the server did not accept these structures. For each tool the whisker plot indicates the median (white dot) and the first quartiles (black box), kernel density estimation of all data points shown as violin plot with minima and maxima indicated by whiskers. b Precision and recall for the test set and 100 randomly drawn structures (except 93 for NI, 68 for CU, 57 for FE2, CO for 30) for other transition (blue), earth-alkali (pink) and alkali (brown) metals. Metal1D as light squares, Metal3D as circles. Threshold used for Metal3D p = 0.75, threshold used for Metal1D t = 0.75.
Fig. 4 Mean absolute deviation of correctly predicted sites and selectivity for other ions. a For all sites where predicted and experimental location are available (true positives in Fig. 3) we compute the mean absolute deviation (MAD) using Metal1D, Metal3D, BioMetAll, MIB and MIB2 on the test set. Some tools predict multiple locations (n) per site (TP). *For MIB and MIB2 we used 2 structures less because the server did not accept these structures. For each tool the whisker plot indicates the median (white dot) and the first quartiles (black box), kernel density estimation of all data points shown as violin plot with minima and maxima indicated by whiskers. b Precision and recall for the test set and 100 randomly drawn structures (except 93 for NI, 68 for CU, 57 for FE2, CO for 30) for other transition (blue), earth-alkali (pink) and alkali (brown) metals. Metal1D as light squares, Metal3D as circles. Threshold used for Metal3D p = 0.75, threshold used for Metal1D t = 0.75.CC BY · PubMed Central
Fig. 5 Annotation of AlphaFold2 structures. a Predicted metal binding sites (a, b) from Metal3D, respectively AlphaFill compared to the experimentally found zinc positions for Uniprot O95630. Metal3D places the metal with high accuracy even if sidechains are not perfectly predicted by AlphaFold for site 1. b Palmitoyltransferase ZDHHC23 (Uniprot Q8IYP9) and ZDHHC15 (Uniprot Q96MV8). AlphaFill can only place the metal for ZDHHC15 because sequence identity for ZDHHC23 is only 24%. Probability isosurfaces from Metal3D for both structures at p = 0.6, colored in gray.
Fig. 5 Annotation of AlphaFold2 structures. a Predicted metal binding sites (a, b) from Metal3D, respectively AlphaFill compared to the experimentally found zinc positions for Uniprot O95630. Metal3D places the metal with high accuracy even if sidechains are not perfectly predicted by AlphaFold for site 1. b Palmitoyltransferase ZDHHC23 (Uniprot Q8IYP9) and ZDHHC15 (Uniprot Q96MV8). AlphaFill can only place the metal for ZDHHC15 because sequence identity for ZDHHC23 is only 24%. Probability isosurfaces from Metal3D for both structures at p = 0.6, colored in gray.CC BY · PubMed Central
Fig. 6 Protein design application. Experimentally measured Kd values65 -- 69 for 1st and 2nd shell active site mutants of HCA2 and predicted max probability for zinc using Metal3D.
Fig. 6 Protein design application. Experimentally measured Kd values65 -- 69 for 1st and 2nd shell active site mutants of HCA2 and predicted max probability for zinc using Metal3D.CC BY · PubMed Central
Fig. 7 Metal1D method. a Coordination environment extraction from LINK records of PDB files and probability map generation b Site prediction, showing examples for a Zn2+ binding site, correctly predicted, and unlikely Zn2+ binding site, discarded in the putative site scoring step because of low probability of the metal site, and an amino acid for which Zn2+ binding is impossible, discarded in the residue scoring step because of low score of the amino acid.
Fig. 7 Metal1D method. a Coordination environment extraction from LINK records of PDB files and probability map generation b Site prediction, showing examples for a Zn2+ binding site, correctly predicted, and unlikely Zn2+ binding site, discarded in the putative site scoring step because of low probability of the metal site, and an amino acid for which Zn2+ binding is impossible, discarded in the residue scoring step because of low score of the amino acid.CC BY · PubMed Central
Fig. 8 Voxelization of an environment. A Cα centered box is voxelized using moleculekit library. For each atom an atom centered pair correlation function is used to assign voxel values in 8 different channels (see Table S5) for atom types.
Fig. 8 Voxelization of an environment. A Cα centered box is voxelized using moleculekit library. For each atom an atom centered pair correlation function is used to assign voxel values in 8 different channels (see Table S5) for atom types.CC BY · PubMed Central
Fig. 9 Sketch of model. Residue centered zinc densities are predicted based on input environment. Number of layers not to scale. Blue tensor sizes indicate number of channels and size of grid (always 32 × 32 × 32) Convolutional filters (red) are trained and extract information (3 × 3 × 3 convolutions) or aggregate information (16 × 16 × 16 convolution).
Fig. 9 Sketch of model. Residue centered zinc densities are predicted based on input environment. Number of layers not to scale. Blue tensor sizes indicate number of channels and size of grid (always 32 × 32 × 32) Convolutional filters (red) are trained and extract information (3 × 3 × 3 convolutions) or aggregate information (16 × 16 × 16 convolution).CC BY · PubMed Central
Fig. 10 Grid averaging and metal ion placement. A bounding box for the global grid is defined based on all predictions and residue probability maps are aggregated using KD-Tree search. Ions are placed after AgglomerativeClustering and taking the weighted average of all voxels in a cluster. Probability is the maximum probability of the cluster.
Fig. 10 Grid averaging and metal ion placement. A bounding box for the global grid is defined based on all predictions and residue probability maps are aggregated using KD-Tree search. Ions are placed after AgglomerativeClustering and taking the weighted average of all voxels in a cluster. Probability is the maximum probability of the cluster.CC BY · PubMed Central

Verification

Claims the paper makes

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

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

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: 79

  1. 10.1021/cr400458x · DOI · Paper page
  2. 10.1093/protein/gzw026 · DOI
  3. 10.1007/s00775-008-0404-5 · DOI · Paper page
  4. 2019, Predicting Disease-Associated Mutation of Metal-Binding Sites in Proteins Using a Deep Learning Approach, Nat Mach Intell, 1, 561, 10.1038/s42256-019-0119-z · DOI
  5. 10.1126/science.aau3744 · DOI · Paper page
  6. 10.1038/nature17968 · DOI
  7. 2021, De Novo Metalloprotein Design, Nat Rev Chem, 6, 31, 10.1038/s41570-021-00339-5 · DOI · Paper page
  8. 10.1021/bi201881p · DOI · Paper page
  9. 2015, Enzyme Repurposing of a Hydrolase as an Emergent Peroxidase Upon Metal Binding, Chem. Sci., 6, 4060, 10.1039/C5SC01065A · DOI
  10. 2012, Metal-Directed, Chemically Tunable Assembly of One-, Two- and Three-Dimensional Crystalline Protein Arrays, Nature Chem, 4, 375, 10.1038/nchem.1290 · DOI · Paper page
  11. 2011, Metal-Mediated Affinity and Orientation Specificity in a Computationally Designed Protein Homodimer, J. Am. Chem. Soc., 134, 375
  12. 10.1021/ar900273t · DOI · Paper page
  13. 2020, Metal-Templated Design of Chemically Switchable Protein Assemblies with High◻Affinity Coordination Sites, Angew. Chem. Int. Ed., 59, 21940, 10.1002/anie.202009226 · DOI
  14. 2011, Hydrolytic Catalysis and Structural Stabilization in a Designed Metalloprotein, Nature Chem, 4, 118
  15. 2020, Thermodynamics of Transition Metal Ion Binding to Proteins, J. Am. Chem. Soc., 142, 6365, 10.1021/jacs.0c01329 · DOI
  16. 10.1021/cr500628b · DOI · Paper page
  17. 2021, Multiscale Workflow for Modeling Ligand Complexes of Zinc Metalloproteins, J. Chem. Inf. Model., 61, 5658, 10.1021/acs.jcim.1c01109 · DOI
  18. 2018, Genetic Algorithm Based Design and Experimental Characterization of a Highly Thermostable Metalloprotein, J. Am. Chem. Soc., 140, 4517, 10.1021/jacs.7b10660 · DOI
  19. https://doi.org/10.3389/fchem.2021.692200. · DOI
  20. Passerini, A. ; Andreini, C. ; Menchetti, S. ; Rosato, A. ; Frasconi, P. Predicting Zinc Binding at the Proteome Level. BMC Bioinformatics 2007, 8 (1). https://doi.org/10.1186/1471-2105-8-39. · DOI
  21. 10.1093/bioinformatics/btw396 · DOI · Paper page
  22. 10.1021/acs.jcim.6b00407 · DOI · Paper page
  23. Hekkelman, M. L. ; de Vries, I. ; Joosten, R. P. ; Perrakis, A. AlphaFill: Enriching the AlphaFold Models with Ligands and Co-Factors, 2021. https://doi.org/10.1101/2021.11.26.470110. · DOI · Paper page
  24. 2010, FINDSITE-Metal: Integrating Evolutionary Information and Machine Learning for Structure-Based Metal-Binding Site Prediction at the Proteome Level, Proteins, 79, 735
  25. 2020, BioMetAll: Identifying Metal-Binding Sites in Proteins from Backbone Preorganization, J. Chem. Inf. Model., 61, 311
  26. 2019, Prediction of Protein Metal Binding Sites Using Deep Neural Networks, Mol. Inf., 38, 1800169, 10.1002/minf.201800169 · DOI
  27. 10.1126/science.abj8754 · doi:10.1126/science.abj8754 · Paper page
  28. Torng, W. ; Altman, R. B. 3D Deep Convolutional Neural Networks for Amino Acid Environment Similarity Analysis. BMC Bioinformatics 2017, 18 (1). https://doi.org/10.1186/s12859-017-1702-0. · DOI · Paper page
  29. 2020, Discovery of Novel Gain-of-Function Mutations Guided by Structure-Based Deep Learning, ACS Synth. Biol., 9, 2927, 10.1021/acssynbio.0c00345 · DOI · Paper page
  30. Anand, N. ; Eguchi, R. ; Mathews, I. I. ; Perez, C. P. ; Derry, A. ; Altman, R. B. ; Huang, P.-S. Protein Sequence Design with a Learned Potential. Nat Commun 2022, 13 (1). https://doi.org/10.1038/s41467-022-28313-9. · DOI · Paper page
  31. 2018, High Precision Protein Functional Site Detection Using 3D Convolutional Neural Networks, Bioinformatics, 35, 1503
  32. Feehan, R. ; Franklin, M. W. ; Slusky, J. S. G. Machine Learning Differentiates Enzymatic and Non-Enzymatic Metals in Proteins. Nat Commun 2021, 12 (1). https://doi.org/10.1038/s41467-021-24070-3. · DOI · Paper page
  33. Renaud, N. ; Geng, C. ; Georgievska, S. ; Ambrosetti, F. ; Ridder, L. ; Marzella, D. F. ; Réau, M. F. ; Bonvin, A. M. J. J. ; Xue, L. C. DeepRank: A Deep Learning Framework for Data Mining 3D Protein-Protein Interfaces. Nat Commun 2021, 12 (1). https://doi.org/10.1038/s41467-021-27396-0. · DOI · Paper page
  34. 2019, Deciphering Interaction Fingerprints from Protein Molecular Surfaces Using Geometric Deep Learning, Nat Methods, 17, 184
  35. 10.1093/bioinformatics/btx350 · DOI · Paper page
  36. 2018, LigVoxel: Inpainting Binding Pockets Using 3D-Convolutional Neural Networks, Bioinformatics, 35, 243
  37. Stärk, H. ; Ganea, O.-E. ; Pattanaik, L. ; Barzilay, R. ; Jaakkola, T. EquiBind: Geometric Deep Learning for Drug Binding Structure Prediction. arXiv 2022. https://doi.org/10.48550/arxiv.2202.05146. · arXiv · doi:10.48550/arxiv.2202.05146 · Paper page
  38. 2022, GalaxyWater-CNN: Prediction of Water Positions on the Protein Structure by a 3D-Convolutional Neural Network, J. Chem. Inf. Model., 62, 3157, 10.1021/acs.jcim.2c00306 · DOI
  39. Li, B. ; Yang, Y. T. ; Capra, J. A. ; Gerstein, M. B. Predicting Changes in Protein Thermodynamic Stability Upon Point Mutation with Deep 3D Convolutional Neural Networks. PLoS Comput Biol 2020, 16 (11), e1008291. https://doi.org/10.1371/journal.pcbi.1008291. · DOI · Paper page
  40. 10.1371/journal.pone.0039252 · DOI · Paper page
  41. 10.1021/ic401072d · DOI · Paper page
  42. Minasov, G. ; Vorontsov, I. I. ; Shuvalova, L. ; Brunzelle, J. S. ; Kiryukhina, O. ; Collart, F. R. ; Joachimiak, A. ; Anderson, W. F. ;. Crystal Structure of Unknown Conserved ybaA Protein from Shigella Flexneri, 2007. https://doi.org/10.2210/pdb2okq/pdb. · doi:10.2210/pdb2okq/pdb
  43. Itoh, T. ; Nakagawa, E. ; Yoda, M. ; Nakaichi, A. ; Hibi, T. ; Kimoto, H. Crystal Structure of Alginate Lyase from Paenibacillus Sp. Str. FPU-7, 2019. https://doi.org/10.2210/pdb6kfn/pdb. · DOI
  44. Adams, C. M. ; Eckenroth, B. E. ; Doublie, S. Structure of the Clostridium Perfringens CspB Protease, 2013. https://doi.org/10.2210/pdb4i0w/pdb. · doi:10.2210/pdb4i0w/pdb
  45. Davies, C. W. ; Das, C. The Crystal Structure of a E280A Mutant of the Catalytic Domain of AMSH, 2011. https://doi.org/10.2210/pdb3rzv/pdb. · DOI
  46. Rana, M. S. ; Kumar, P. ; Lee, C.-J. ; Verardi, R. ; Rajashankar, K. R. ; Banerjee, A. Fatty Acyl Recognition and Transfer by an Integral Membrane S-Acyltransferase. Science 2018, 359 (6372). https://doi.org/10.1126/science.aao6326. · DOI · Paper page
  47. Hakansson, K. ; Carlsson, M. ; Svensson, L. A. ; Liljas, A. STRUCTURE OF NATIVE AND APO CARBONIC ANHYDRASE II AND SOME OF ITS ANION-LIGAND COMPLEXES, 1993. https://doi.org/10.2210/pdb2cba/pdb. · DOI
  48. 10.1016/0003-2697(85)90409-9 · DOI · Paper page
  49. 2000, Successful Molecular Dynamics Simulation of the Zinc-Bound Farnesyltransferase Using the Cationic Dummy Atom Approach, Protein Sci, 9, 1857
  50. 10.1021/bi00255a003 · DOI
  51. 10.1021/ja00079a046 · DOI
  52. 1993, Structure of an Engineered His3 Cys Zinc Binding Site in Human Carbonic Anhydrase II, Biochemistry, 32, 9901, 10.1021/bi00089a005 · DOI
  53. 10.1073/pnas.92.11.5017 · DOI
  54. 10.1021/bi9526692 · DOI
  55. Savage, H. ; Wlodawer, A. Determination of Water Structure Around Biomolecules Using X-Ray and Neutron Diffraction Methods. Methods Enzymol 1986, 127, 162–183. https://doi.org/10.1016/0076-6879(86)27014-7. · DOI
  56. 10.1002/prot.25081 · DOI
  57. 2017, Waterdock 2.0: Water Placement Prediction for Holo-Structures with a Pymol Plugin, PLoS ONE, 12, e0172743, 10.1371/journal.pone.0172743 · DOI
  58. Satorras, V. G. ; Hoogeboom, E. ; Welling, M. E(n) Equivariant Graph Neural Networks. arXiv 2021. https://doi.org/10.48550/arxiv.2102.09844. · arXiv · doi:10.48550/arxiv.2102.09844 · Paper page
  59. Gligorijević, V. ; Berenberg, D. ; Ra, S. ; Watkins, A. ; Kelow, S. ; Cho, K. ; Bonneau, R. Function-Guided Protein Design by Deep Manifold Sampling, 2021. https://doi.org/10.1101/2021.12.22.473759. · DOI
  60. Greener, J. G. ; Moffat, L. ; Jones, D. T. Design of Metalloproteins and Novel Protein Folds Using Variational Autoencoders. Sci Rep 2018, 8 (1). https://doi.org/10.1038/s41598-018-34533-1. · DOI · Paper page
  61. 2012, Revisiting Zinc Coordination in Human Carbonic Anhydrase II, Inorg. Chem., 51, 11098, 10.1021/ic301645j · DOI
  62. 10.1126/science.8346440 · DOI · Paper page
  63. 10.1126/science.1648261 · DOI · Paper page
  64. 10.1038/nsb723 · DOI
  65. 2021, Highly Accurate Protein Structure Prediction for the Human Proteome, Nature, 596, 590, 10.1038/s41586-021-03828-1 · DOI · Paper page
  66. 10.1093/nar/28.1.235 · DOI · Paper page
  67. 10.1038/nbt.3988 · DOI · Paper page
  68. Barber-Zucker, S. ; Shaanan, B. ; Zarivach, R. Transition Metal Binding Selectivity in Proteins and Its Correlation with the Phylogenomic Classification of the Cation Diffusion Facilitator Protein Family. Sci Rep 2017, 7 (1). https://doi.org/10.1038/s41598-017-16777-5. · DOI
  69. 2017, BioPandas: Working with Molecular Structures in Pandas DataFrames, JOSS, 2, 279, 10.21105/joss.00279 · DOI · Paper page
  70. 10.1021/acs.jctc.6b00049 · DOI · Paper page
  71. Moritz, P. ; Nishihara, R. ; Wang, S. ; Tumanov, A. ; Liaw, R. ; Liang, E. ; Elibol, M. ; Yang, Z. ; Paul, W. ; Jordan, M. I. ; Stoica, I. Ray: A Distributed Framework for Emerging AI Applications. arXiv 2017. https://doi.org/10.48550/arxiv.1712.05889. · arXiv · doi:10.48550/arxiv.1712.05889 · Paper page
  72. Paszke, A. ; Gross, S. ; Massa, F. ; Lerer, A. ; Bradbury, J. ; Chanan, G. ; Killeen, T. ; Lin, Z. ; Gimelshein, N. ; Antiga, L. ; Desmaison, A. ; Köpf, A. ; Yang, E. ; DeVito, Z. ; Raison, M. ; Tejani, A. ; Chilamkurthy, S. ; Steiner, B. ; Fang, L. ; Bai, J. ; Chintala, S. PyTorch: An Imperative Style, High-Performance Deep Learning Library. arXiv 2019. https://doi.org/10.48550/arxiv.1912.01703. · DOI · Paper page
  73. 2005, A Tutorial on the Cross-Entropy Method, Ann Oper Res, 134, 19, 10.1007/s10479-005-5724-z · DOI · Paper page
  74. 10.1038/s41592-019-0686-2 · DOI · Paper page
  75. Pedregosa, F. ; Varoquaux, G. ; Gramfort, A. ; Michel, V. ; Thirion, B. ; Grisel, O. ; Blondel, M. ; Müller, A. ; Nothman, J. ; Louppe, G. ; Prettenhofer, P. ; Weiss, R. ; Dubourg, V. ; Vanderplas, J. ; Passos, A. ; Cournapeau, D. ; Brucher, M. ; Perrot, M. ; Duchesnay, É. Scikit-Learn: Machine Learning in Python. arXiv 2012. https://doi.org/10.48550/arxiv.1201.0490. · DOI · Paper page
  76. 10.1016/0263-7855(96)00018-5 · DOI · Paper page
  77. 2014, 3Dmol.js: Molecular Visualization with WebGL, Bioinformatics, 31, 1322
  78. 10.1016/0022-2836(92)90531-N · DOI · Paper page
  79. 10.1002/1097-0134(20000815)40:3<389::AID-PROT50>3.0.CO;2-2 · DOI · Paper page

Metadata

Authors
Dürr, S. L., Levy, A., & Rothlisberger, U.
Affiliations
École Polytechnique Fédérale de Lausanne (CH)
Venue
Nature Communications
Publisher
openRxiv
Published
2022-08-22
Volume
14
Issue
1
Pages
2713-2713
ISSN
2041-1723 (ISSN-L)
OpenAlex
W4376225285
Cited by
88
References
1
Citations by year
2026: 35 · 2025: 30 · 2024: 19 · 2023: 4
FWCI
7.84
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Nature Communications (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Infoscience (Ecole Polytechnique Fédérale de Lausanne) (submittedVersion, cc-by-nc-nd)
Funders
  • National Science Foundation (200020)
  • Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (185092, 200020)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Structural Biology › Enzyme Structure and Function
Keywords
zinc ions, metalloproteins, protein structure, 3D convolutional neural networks, geometric criterion, protein design, cofactors
MeSH terms
Deep Learning, Binding Sites, Ions, Metalloproteins, Metals, Zinc

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

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.