Hierarchical, rotation‐equivariant neural networks to select structural models of protein complexes
Open accessThe DOI matches its Crossref record; 10 of the 10 references it cites were found in a registry or an index.
Abstract
Predicting the structure of multi‐protein complexes is a grand challenge in biochemistry, with major implications for basic science and drug discovery. Computational structure prediction methods generally leverage predefined structural features to distinguish accurate structural models from less accurate ones. This raises the question of whether it is possible to learn characteristics of accurate models directly from atomic coordinates of protein complexes, with no prior assumptions. Here we introduce a machine learning method that learns directly from the 3D positions of all atoms to identify accurate models of protein complexes, without using any precomputed physics‐inspired or statistical terms. Our neural network architecture combines multiple ingredients that together enable end‐to‐end learning from molecular structures containing tens of thousands of atoms: a point‐based representation of atoms, equivariance with respect to rotation and translation, local convolutions, and hierarchical subsampling operations. When used in combination with previously developed scoring functions, our network substantially improves the identification of accurate structural models among a large set of possible models. Our network can also be used to predict the accuracy of a given structural model in absolute terms. The architecture we present is readily applicable to other tasks involving learning on 3D structures of large atomic systems.
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.
Integrity signals
Registry
Cites
Statistics
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.
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
Generalized Neural-Network Representation of High-Dimensional Potential-Energy Surfaces
Behler, J., & Parrinello, M. (2007) · Physical Review Letters · doi:10.1103/physrevlett.98.146401 · Cited by 5,010 · cited 2×
“To account for the fact that the laws of physics that govern inter-and intramolecular interactions are invariant to translations and rotations, Behler and Parrinello 30 developed basic radial and angular symmetry functions in order to model the potential energy of small molecules through neural networks.”
BackgroundANI-1: an extensible neural network potential with DFT accuracy at force field computational cost
Smith, J. S., Isayev, O., & Roitberg, A. E. (2017) · Chemical Science · doi:10.1039/c6sc05720a · Cited by 1,921 · cited 2×
“These were further extended in SchNet 31 and ANI-1.”
BackgroundHorovod: fast and easy distributed deep learning in TensorFlow
Sergeev, A., & Del Balso, M. (2018) · arXiv · doi:10.48550/arxiv.1802.05799 · Cited by 1,620
ZDOCK: An initial‐stage protein‐docking algorithm
Chen, R., Li, L., & Weng, Z. (2003) · Proteins: Structure, Function, and Bioinformatics · doi:10.1002/prot.10389 · Cited by 1,393 · cited 1×
“Despite the availability of many software packages designed for this purpose, 3-11 identifying accurate models out of a large set of candidate models has proven to be a major bottleneck for successful protein docking.”
MethodTensor field networks: Rotation- and translation-equivariant neural networks for 3D point clouds
Thomas, N., Smidt, T., Kearnes, S., Yang, L., Li, L., Kohlhoff, K., & Riley, P. (2018) · arXiv · doi:10.48550/arxiv.1802.08219 · Cited by 1,303 · cited 5×
“Concretely, we can represent inputs as V am (the coordinates and other features associated with each point), where a indexes the points and m is an index over the group representation (and in general will be broken up into multiple indices based on the internal structure of the representation, as described in Thomas e…”
MethodInfluentialSelf-citationPrediction Errors of Molecular Machine Learning Models Lower than Hybrid DFT Error
Faber, F. A., Hutchison, L., Huang, B., Gilmer, J., Schoenholz, S. S., Dahl, G. E., Vinyals, O., Kearnes, S., Riley, P. F., & von Lilienfeld, O. A. (2017) · Journal of Chemical Theory and Computation · doi:10.1021/acs.jctc.7b00577 · Cited by 651
Vreven, T., Moal, I. H., Vangone, A., Pierce, B. G., Kastritis, P. L., Torchala, M., Chaleil, R., Jiménez-García, B., Bates, P. A., Fernandez-Recio, J., et al. (2015) · Journal of Molecular Biology · doi:10.1016/j.jmb.2015.07.016 · Cited by 424 · cited 4×
“DB4 and its successor docking benchmark 5 (DB5) 39 are widely used datasets specifically curated to benchmark protein docking methods as well as scoring functions used for protein docking.”
BackgroundMethodInfluentialProtein-Protein Docking: From Interaction to Interactome
Vakser, I. A. (2014) · Biophysical Journal · doi:10.1016/j.bpj.2014.08.033 · Cited by 318 · cited 1×
“Determining structures of protein complexes experimentally is often difficult and time-consuming, placing a premium on computational structure prediction.(2) Unfortunately, computational structure prediction for protein complexes has also proven difficult—much more so than for individual proteins.”
Background3D deep convolutional neural networks for amino acid environment similarity analysis
Torng, W., & Altman, R. B. (2017) · BMC Bioinformatics · doi:10.1186/s12859-017-1702-0 · Cited by 142
Protein-protein and peptide-protein docking and refinement using ATTRACT in CAPRI
Schindler, C. E. M., Chauvot de Beauchêne, I., de Vries, S. J., & Zacharias, M. (2017) · Proteins Structure Function and Bioinformatics · doi:10.1002/prot.25196 · Cited by 20 · cited 3×
“For this second dataset, we use the homology models of the individual proteins to generate models of protein complexes with ATTRACT.”
MethodInfluential
Cited by
Geometric deep learning of RNA structure
Townshend, R. J. L., Eismann, S., Watkins, A. M., Rangan, R., Karelina, M., Das, R., & Dror, R. O. (2021) · Science · doi:10.1126/science.abe5650 · Cited by 414
Self-citationStructure-based drug design with geometric deep learning
Isert, C., Atz, K., & Schneider, G. (2023) · Current Opinion in Structural Biology · doi:10.1016/j.sbi.2023.102548 · Cited by 182
DeepRank: a deep learning framework for data mining 3D protein-protein interfaces
Renaud, N., Geng, C., Georgievska, S., Ambrosetti, F., Ridder, L., Marzella, D. F., Réau, M. F., Bonvin, A. M. J. J., & Xue, L. C. (2021) · Nature Communications · doi:10.1038/s41467-021-27396-0 · Cited by 125
Determining structures of RNA conformers using AFM and deep neural networks
Degenhardt, M. F. S., Degenhardt, H. F., Bhandari, Y. R., Lee, Y.-T., Ding, J., Yu, P., Heinz, W. F., Stagno, J. R., Schwieters, C. D., Watts, N. R., et al. (2025) · Nature · doi:10.1038/s41586-024-07559-x · Cited by 56
Protein sequence‐to‐structure learning: Is this the end(‐to‐end revolution)?
Laine, E., Eismann, S., Elofsson, A., & Grudinin, S. (2021) · Proteins: Structure, Function, and Bioinformatics · doi:10.1002/prot.26235 · Cited by 44
Self-citationDeep Local Analysis evaluates protein docking conformations with locally oriented cubes
Mohseni Behbahani, Y., Crouzet, S., Laine, E., & Carbone, A. (2022) · Bioinformatics · doi:10.1093/bioinformatics/btac551 · Cited by 16
Metadata
- Authors
- Eismann, S., Townshend, R. J. L., Thomas, N., Jagota, M., Jing, B., & Dror, R. O.
- Venue
- Proteins: Structure, Function, and Bioinformatics
- Publisher
- Wiley
- Published
- 2020-12-31
- Volume
- 89
- Issue
- 5
- Pages
- 493-501
- ISSN
- 0887-3585 (ISSN-L)
- 1097-0134
- OpenAlex
- W3110563343
- Cited by
- 55
- References
- 10
- FWCI
- 2.70
- Type
- journal article
- Language
- en
- Open access
- green
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.