Predictions from Deep Learning Propose Substantial Protein-Carbohydrate Interplay
Open accessPreprint — published as 10.1073/pnas.2523342123 · Paper page
Abstract
It is a grand challenge to identify all the protein – carbohydrate interactions in an organism. Direct experiments would require extensive libraries of glycans to definitively distinguish binding from non-binding proteins. Computational screening of proteins for carbohydrate-binding provides an attractive and ultimately testable alternative. Recent computational techniques have focused primarily on which protein residues interact with carbohydrates or which carbohydrate species a protein binds to. Current estimates label 1.5 to 5% of proteins as carbohydrate-binding proteins; however, 50-70% of proteins are known to be glycosylated, suggesting a potential wealth of proteins that bind to carbohydrates. We therefore developed a novel dataset and neural network architecture, named P rotein i nteraction of Ca rbohydrates P redictor (PiCAP), to predict whether a protein non-covalently binds to a carbohydrate. We trained PiCAP on a dataset of known carbohydrate binders, and we selected proteins that we identified as likely not to bind carbohydrates, including DNA-binding transcription factors, cytoskeletal components, selected antibodies, and selected small-molecule-binding proteins. PiCAP achieves a 90% balanced accuracy on protein-level predictions of carbohydrate binding/non-binding. Using the same dataset, we developed a model named Ca rbohydrate P rotein S ite I denti f ier 2 (CAPSIF2) to predict protein residues that interact non-covalently with carbohydrates. CAPSIF2 achieves a Dice coefficient of 0.57 on residue-level predictions on our independent test dataset, outcompeting all previous models for this task. To demonstrate the biological applicability of PiCAP and CAPSIF2, we investigated cell surface proteins of human neural cells and further predicted the likelihood of three proteomes, notably E. coli, M. musculus , and H. sapiens , to bind to carbohydrates. PiCAP predicts that approximately 35-40% of proteins in these proteomes bind carbohydrates, indicating a substantial interplay of protein-carbohydrate interactions for cellular functionality. Significance Statement: The totality of carbohydrate-protein interactions remains elusive, in part due to the inability to test proteomes versus glycomes in a high throughput manner. Here we show the first high-throughput methodology to predict protein-carbohydrate interactions at proteomic scales by using structural and sequence information. This information will allow scientists to target predicted protein-carbohydrate interactions to better determine how the elusive carbohydrate biomolecules play roles in all cellular functions.
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.
Integrity signals
Registry
Cites
Statistics
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.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
Highly accurate protein structure prediction with AlphaFold
Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Žídek, A., Potapenko, A., et al. (2021) · Nature · doi:10.1038/s41586-021-03819-2 · Cited by 43,764 · see citing works
Accurate structure prediction of biomolecular interactions with AlphaFold 3
Abramson, J., Adler, J., Dunger, J., Evans, R., Green, T., Pritzel, A., Ronneberger, O., Willmore, L., Ballard, A. J., Bambrick, J., et al. (2024) · Nature · doi:10.1038/s41586-024-07487-w · Cited by 15,861 · cited 2× · see citing works
“…an initial hypothesis to feed GlycanDock 42 , or directly de novo with programs like DiffDock 13 , RosettaFold-All Atom (RF-AA) 43 , AlphaFold3 11 , or Boltz-1 44 (although there are currently no validation studies testing whether these methods provide high accuracy on carbohydrate-specific…”
ColabFold - Making protein folding accessible to all
Mirdita, M., Schütze, K., Moriwaki, Y., Heo, L., Ovchinnikov, S., & Steinegger, M. (2021) · doi:10.1101/2021.08.15.456425 · Cited by 9,391 · see citing works
UniProt: the Universal Protein Knowledgebase in 2023
The UniProt Consortium, Bateman, A., Martin, M.-J., Orchard, S., Magrane, M., Ahmad, S., Alpi, E., Bowler-Barnett, E. H., Britto, R., Bye-A-Jee, H., et al. (2023) · Nucleic Acids Research · doi:10.1093/nar/gkac1052 · Cited by 7,210 · see citing works
Evolutionary-scale prediction of atomic level protein structure with a language model
Lin, Z., Akin, H., Rao, R., Hie, B., Zhu, Z., Lu, W., Smetanin, N., Verkuil, R., Kabeli, O., Shmueli, Y., et al. (2022) · doi:10.1101/2022.07.20.500902 · Cited by 5,523 · see citing works
A classification of glycosyl hydrolases based on amino acid sequence similarities.
Henrissat, B. (1991) · Biochemical Journal · doi:10.1042/bj2800309 · Cited by 3,188 · cited 1× · see citing works
“For comparison, the total number of lectins identified by LectomeXplore is 230, or 1.1% of the UniProt reference proteome 22 , and the number known by CaZy is 349, or 1.7% 21 .”
InfluentialVaradi, M., Bertoni, D., Magana, P., Paramval, U., Pidruchna, I., Radhakrishnan, M., Tsenkov, M., Nair, S., Mirdita, M., Yeo, J., et al. (2024) · Nucleic Acids Research · doi:10.1093/nar/gkad1011 · Cited by 2,052 · cited 2× · see citing works
“We chose three model organisms from the AF2 proteome datasets 12 , E. coli , M. musculus, and H. sapiens .”
The CRAPome: a contaminant repository for affinity purification–mass spectrometry data
Mellacheruvu, D., Wright, Z., Couzens, A. L., Lambert, J.-P., St-Denis, N. A., Li, T., Miteva, Y. V., Hauri, S., Sardiu, M. E., Low, T. Y., et al. (2013) · Nature Methods · doi:10.1038/nmeth.2557 · Cited by 1,686 · see citing works
Large language models generate functional protein sequences across diverse families
Madani, A., Krause, B., Greene, E. R., Subramanian, S., Mohr, B. P., Holton, J. M., Olmos, J. L., Xiong, C., Sun, Z. Z., Socher, R., et al. (2023) · Nature Biotechnology · doi:10.1038/s41587-022-01618-2 · Cited by 1,078 · cited 1× · see citing works
“20 Further, we included the computationally designed and experimentally viable lysozymes from ProGen, 24 with structures predicted by the Colab distribution of AlphaFold2.”
Generalized biomolecular modeling and design with RoseTTAFold All-Atom
Krishna, R., Wang, J., Ahern, W., Sturmfels, P., Venkatesh, P., Kalvet, I., Lee, G. R., Morey-Burrows, F. S., Anishchenko, I., Humphreys, I. R., et al. (2024) · Science · doi:10.1126/science.adl2528 · Cited by 978 · cited 2× · see citing works
“Another step would a high throughput computational docking of those carbohydrate species to the identified proteins, using CAPSIF2 or PesTo-Carbs 17 or DeepGlycanSite 18 to identify an initial hypothesis to feed GlycanDock 42 , or directly de novo with programs like DiffDock 13 , RosettaFold-All Atom (RF-AA) 43 , Alph…”
DiffDock: Diffusion Steps, Twists, and Turns for Molecular DockingPreprint
Corso, G., Stärk, H., Jing, B., Barzilay, R., & Jaakkola, T. (2022) · International Conference on Learning Representations · doi:10.48550/arxiv.2210.01776 · Cited by 888 · cited 2× · see citing works
“Another step would a high throughput computational docking of those carbohydrate species to the identified proteins, using CAPSIF2 or PesTo-Carbs 17 or DeepGlycanSite 18 to identify an initial hypothesis to feed GlycanDock 42 , or directly de novo with programs like DiffDock 13 , RosettaFold-All Atom (RF-AA) 43 , Alph…”
PyRosetta: a script-based interface for implementing molecular modeling algorithms using Rosetta
Chaudhury, S., Lyskov, S., & Gray, J. J. (2010) · Bioinform. · doi:10.1093/bioinformatics/btq007 · Cited by 772 · see citing works
Self-citationSAbDab: the structural antibody database
Dunbar, J., Krawczyk, K., Leem, J., Baker, T., Fuchs, A., Georges, G., Shi, J., & Deane, C. M. (2014) · Nucleic Acids Research · doi:10.1093/nar/gkt1043 · Cited by 674 · see citing works
Boltz-1 Democratizing Biomolecular Interaction ModelingPreprint
Wohlwend, J., Corso, G., Passaro, S., Getz, N., Reveiz, M., Leidal, K., Swiderski, W., Atkinson, L., Portnoi, T., Chinn, I., et al. (2024) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2024.11.19.624167 · Cited by 535 · see citing works
Forging the basis for developing protein-ligand interaction scoring functions.
Liu, Z., Su, M., Han, L., Liu, J., Yang, Q., Li, Y., & Wang, R. (2017) · Accounts of Chemical Research · doi:10.1021/acs.accounts.6b00491 · Cited by 493 · see citing works
E(n) Equivariant Graph Neural NetworksPreprint
Satorras, V. G., Hoogeboom, E., & Welling, M. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.09844 · Cited by 481 · see citing works
A highlight on Sonic hedgehog pathway
Carballo, G. B., Honorato, J. R., de Lopes, G. P. F., & Spohr, T. C. L. D. S. E. (2018) · Cell Communication and Signaling · doi:10.1186/s12964-018-0220-7 · Cited by 460 · cited 1× · see citing works
“PiCAP predicts sonic hedgehog (SHH), a protein implicated in cellular differentiation that can exist in the extracellular space to mediate the anterior-posterior axis, as a carbohydrate binding protein 34 .”
Structure of the bacterial ribosome at 2 Å resolution
Watson, Z. L., Ward, F. R., Méheust, R., Ad, O., Schepartz, A., Banfield, J. F., & Cate, J. H. (2020) · eLife · doi:10.7554/elife.60482 · Cited by 263 · see citing works
Alfaris, N., Waldrop, S., Johnson, V., Boaventura, B., Kendrick, K., & Stanford, F. C. (2024) · EClinicalMedicine · doi:10.1016/j.eclinm.2024.102782 · Cited by 224 · see citing works
DNAH6 and Its Interactions with PCD Genes in Heterotaxy and Primary Ciliary Dyskinesia
Li, Y., Yagi, H., Onuoha, E., Damerla, R., Francis, R., Furutani, Y., Tariq, M., King, S., Hendricks, G., Cui, C., et al. (2016) · PLoS Genetics · doi:10.1371/journal.pgen.1005821 · Cited by 117 · see citing works
Deep learning for protein structure prediction and design—progress and applications
Jänes, J., & Beltrão, P. (2024) · Molecular Systems Biology · doi:10.1038/s44320-024-00016-x · Cited by 71 · see citing works
Tan, X., Lu, T., Li, L., Xu, J., Xie, S., Ji, L., Fu, J., Liu, Q., Shen, S., Liu, Y., et al. (2021) · Nature Communications · doi:10.1038/s41467-021-24898-9 · Cited by 71 · cited 1× · see citing works
“Finally, SMAD4 is a transcription factor 33 , which was identified to not interact with gangliosides in all experiments, where PiCAP agrees and predicts the protein as a carbohydrate non-binder.”
The Protein-DNA Interface database
Norambuena, T., & Melo, F. (2010) · BMC Bioinformatics · doi:10.1186/1471-2105-11-262 · Cited by 62 · see citing works
TMEM164 is an acyltransferase that forms ferroptotic C20:4 ether phospholipids
Reed, A., Ware, T. B., Li, H., Bazán, J. F., & Cravatt, B. F. (2023) · Nature Chemical Biology · doi:10.1038/s41589-022-01253-7 · Cited by 62 · cited 1× · see citing works
“TMEM164 is a regulator of ferropoptosis (iron mediated cell death) by an enzyme reaction with polyunsaturated fatty acids (PUFA) and was identified by three experiments 32 .”
LectinOracle – A Generalizable Deep Learning Model for Lectin-Glycan Binding Prediction
Lundstrøm, J., Korhonen, E., Lisacek, F., & Bojar, D. (2021) · doi:10.1101/2021.08.30.458147 · Cited by 50 · cited 1× · see citing works
“Firstly, we encourage the expansion of this work or LectinOracle 40 or GlyNet 41 to predict carbohydrate species to all carbohydrate binding proteins.”
Highly accurate carbohydrate-binding site prediction with DeepGlycanSite
He, X., Zhao, L.-F., Tian, Y.-P., Li, R., Chu, Q., Gu, Z.-Y., Zheng, M.-Y., Wang, Y., Li, S.-N., Jiang, H.-L., et al. (2024) · Nature Communications · doi:10.1038/s41467-024-49516-2 · Cited by 43 · cited 4× · see citing works
“…a high throughput computational docking of those carbohydrate species to the identified proteins, using CAPSIF2 or PesTo-Carbs 17 or DeepGlycanSite 18 to identify an initial hypothesis to feed GlycanDock 42 , or directly de novo with programs like DiffDock 13 , RosettaFold-All Atom (RF-AA) 43 ,…”
InfluentialBonnardel, F., Mariethoz, J., Pérez, S., Imberty, A., & Lisacek, F. (2020) · Nucleic Acids Res. · doi:10.1093/nar/gkaa1019 · Cited by 42 · cited 11× · see citing works
“For comparison, the total number of lectins identified by LectomeXplore is 230, or 1.1% of the UniProt reference proteome 22 , and the number known by CaZy is 349, or 1.7% 21 .”
InfluentialGlyNet: a multi-task neural network for predicting protein–glycan interactions
Carpenter, E. J., Seth, S., Yue, N., Greiner, R., & Derda, R. (2021) · bioRxiv · doi:10.1101/2021.05.28.446094 · Cited by 26 · cited 1×
“Firstly, we encourage the expansion of this work or LectinOracle 40 or GlyNet 41 to predict carbohydrate species to all carbohydrate binding proteins.”
PeSTo-Carbs: Geometric Deep Learning for Prediction of Protein–Carbohydrate Binding Interfaces
Bibekar, P., Krapp, L. F., & Peraro, M. D. (2024) · Journal of Chemical Theory and Computation · doi:10.1021/acs.jctc.3c01145 · Cited by 22 · cited 16× · see citing works
“PesTo-Carbs also leverages a graph neural network approach; however, it operates at an atom-wise level and only pools to the residue level late in the architecture.”
InfluentialNance, M. L., Labonte, J. W., Adolf-Bryfogle, J., & Gray, J. J. (2021) · Journal of Physical Chemistry B · doi:10.1021/acs.jpcb.1c00910 · Cited by 21 · cited 2× · see citing works
“Another step would a high throughput computational docking of those carbohydrate species to the identified proteins, using CAPSIF2 or PesTo-Carbs 17 or DeepGlycanSite 18 to identify an initial hypothesis to feed GlycanDock 42 , or directly de novo with programs like DiffDock 13 , RosettaFold-All Atom (RF-AA) 43 , Alph…”
Self-citationZhang, G.-L., Porter, M. J., Awol, A. K., Orsburn, B., Canner, S. W., Gray, J. J., O'Meally, R. N., Cole, R. N., & Schnaar, R. L. (2024) · Journal of the American Chemical Society · doi:10.1021/jacs.4c03196 · Cited by 20 · cited 7× · see citing works
“23 These works are limited by their scope, requiring either specific protein families or specific carbohydrate species to interact.”
InfluentialSelf-citationLima, G. M., Jame-Chenarboo, Z., Sojitra, M., Sarkar, S., Carpenter, E. J., Yang, C. Y.-L., Schmidt, E. N., Lai, J. Y. Q., Atrazhev, A. M., Yazdan, D., et al. (2024) · Cell chemical biology · doi:10.1016/j.chembiol.2024.09.010 · Cited by 15 · cited 1× · see citing works
“PiCAP results can be further validated by proteomic evaluation by experiments such as the pull-downs from Zhang et al. 23 or liquid glycan arrays 38,39 .”
Recent advances in photoaffinity labeling strategies to capture Glycan-Protein interactions.
Babulic, J. L., González, F. V. D. L., & Capicciotti, C. J. (2024) · Current Opinion in Chemical Biology · doi:10.1016/j.cbpa.2024.102456 · Cited by 15 · see citing works
UniLectin, A One‐Stop‐Shop to Explore and Study Carbohydrate‐Binding Proteins
Imberty, A., Bonnardel, F., & Lisacek, F. (2021) · Current Protocols · doi:10.1002/cpz1.305 · Cited by 15 · cited 1× · see citing works
“19 Recently, DIONYSUS was released detailing an immense set of experimental carbohydrate binding proteins with non-covalently bound carbohydrate as well as glycosylated proteins.”
MetalNet2: an enhanced server for predicting metal-binding sites in proteomes
Zhang, F., Cheng, Y., Xue, B., Gao, Y., Liu, Y., & Wang, C. (2024) · National Science Review · doi:10.1093/nsr/nwae391 · Cited by 12 · see citing works
Mikkelsen, S. A., Vangheluwe, P., & Andersen, J. P. (2018) · Journal of Biological Chemistry · doi:10.1074/jbc.ra117.000941 · Cited by 12 · cited 1× · see citing works
“ATP2A2 has a specific role in ATP-mediated transport of calcium ions and likely little specific affinity for carbohydrates, let alone gangliosides 31 .”
Structure-Based Neural Network Protein-Carbohydrate Interaction Predictions at the Residue LevelPreprint
Canner, S. W., Shanker, S., & Gray, J. J. (2023) · bioRxiv · doi:10.1101/2023.03.14.531382 · Cited by 9 · cited 2× · see citing works
“Most previous work has used single protein chains for protein-carbohydrate predictions 16,17 ; however, many proteins only exist in the context of multiple chains.”
Self-citationMuffels, I., Schene, I., Rehmann, H., Massink, M., van der Wal, M., Bauder, C. A., Labeur, M., Armando, N. G., Lequin, M., Houben, M., et al. (2023) · American Journal of Human Genetics · doi:10.1016/j.ajhg.2022.12.003 · Cited by 9 · see citing works
DIONYSUS: a database of protein–carbohydrate interfaces
Gheeraert, A., Bailly, T., Ren, Y.-N., Hamraoui, A., Te, J., Meersche, Y. V., Cretin, G., Lin, R. L. F., Gelly, J.-C., Perez, S., et al. (2024) · Nucleic Acids Research · doi:10.1093/nar/gkae890 · Cited by 7 · cited 1× · see citing works
“20 Since experimentally solved structures can be difficult to obtain, especially in the presence of a carbohydrate ligand, some datasets of sequences exist that identify carbohydrate binding proteins.”
Walklett, A. J., Flack, E. K. P., Chidwick, H. S., Hatton, N. E., Keenan, T., Budhadev, D., Walton, J., Thomas, G. H., & Fascione, M. (2024) · Angewandte Chemie · doi:10.1002/anie.202318523 · Cited by 6
Lin, C.-L., Carpenter, E. J., Li, T.-R., Ahmed, T., & Derda, R. (2024) · Methods in molecular biology · doi:10.1007/978-1-0716-3798-2_10 · Cited by 5 · cited 1×
“PiCAP results can be further validated by proteomic evaluation by experiments such as the pull-downs from Zhang et al. 23 or liquid glycan arrays 38,39 .”
Luo, Y., & Parmeggiani, F. (2025) · bioRxiv · doi:10.1021/acs.jcim.6c00126 · Cited by 4 · cited 1× · see citing works
“The designed non-binders were created using Rosetta, where the binding pocket itself was designed but the remainder of the protein remained untouched 25 .”
Abdallah, A., Gillon, E., Rannou, P., Imberty, A., & Halila, S. (2024) · Bioconjugate chemistry · doi:10.1021/acs.bioconjchem.4c00224 · Cited by 3 · see citing works
Flagged first, then by citations
2 of the 45 references couldn’t be matched to a known paper.
Cited by
Sorted by citations
Carpenter, E. J., Peng, C., Wang, S.-K., Greiner, R., & Derda, R. (2025) · doi:10.1101/2025.01.21.633632 · Cited by 10 · see citing works
“An interesting example are C-type lectins, it is not known how many of the C-type lectins actually bind to traditional or non-traditional glycans 15,16 .”
Accurate structural modelling of protein-glycan recognition with GlycoWeave
Zhao, H.-F., Zhang, J.-T., Xiao, Y.-F., Qian, Y.-Q., Zhai, S.-L., Jin, Z.-C., Wang, L., Wang, Y.-C., Dong, H., Zhao, Y.-F., et al. (2026) · doi:10.48550/arxiv.2610.37410
MIEN: a deep-learning model for pocket-conditioned glycan pose and conformation discrimination
Huang, X.-R., & Tian, Z.-X. (2026) · bioRxiv · doi:10.64898/2026.10.13.758804
“Protein-level classifiers separate binders from curated non-binders, but they do not resolve the conformation inside a pocket 20 .”
- Authors
- Canner, S. W., Schnaar, R. L., & Gray, J. J.
- Affiliations
- Johns Hopkins University (US)
- Corresponding author
- Gray, Jeffrey J.
- Venue
- Proceedings of the National Academy of Sciences
- Publisher
- openRxiv
- Published
- 2025-03-12
- Volume
- 123
- Issue
- 21
- Pages
- e2523342123-e2523342123
- ISSN
- 0027-8424 (ISSN-L)
- PMC
- PMC11952328
- OpenAlex
- W7161564967
- Cited by
- 4
- References
- 45
- Citations by year
- 2026: 2 · 2025: 2
- FWCI
- 4.10
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- Proceedings of the National Academy of Sciences (publishedVersion, cc-by)
- bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc)
- PubMed (publishedVersion)
- PubMed Central (publishedVersion, cc-by)
- Europe PMC (PubMed Central) (submittedVersion, cc-by-nc)
- PubMed Central (submittedVersion, cc-by-nc)
- Funders
- Johns Hopkins University
- University of Bristol
- National Institutes of Health (1r01gm160587-01)
- National Institute of General Medical Sciences (R01-GM160587, R35GM141881, T32GM135131)
- National Institute of Allergy and Infectious Diseases (R01AI162381)
- Topics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Glycosylation and Glycoproteins Research
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
- Keywords
- carbohydrate-protein interactions, glycans, neural network predictor, protein-protein interaction network, cell surface proteins, extracellular proteins, proteome-wide prediction, growth factor receptors, cell adhesion
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
