Glycans are ubiquitous across all forms of life and are integral to a wide range of biological functions, such as immune response, cell adhesion, and host–pathogen interactions.[1] Their intrinsic complexity arises from the diversity of glycosidic linkages between monosaccharides, the extensive possibilities for branching, the presence of two anomeric configurations, multiple sugar ring conformations and puckering modes, and a wide range of chemical modifications. Current computational approaches, such as molecular docking and molecular dynamics simulations, can generate structural models; however, these predictions are computationally expensive must still be corroborated and refined through empirical data. [2]
Representation of the structural complexity of glycans.
AlphaFold, the AI system that predicts the 3D structures and interactions of proteins, has revolutionized protein structure prediction with remarkable accuracy. Current version of Alphafold 3 allows the modelling of DNA, RNA, small molecules and glycan containing biomolecular complexes too. [3] In its latest version, protein glycosylation was included among PTMs, however the ambiguity in specifying linkages and higher-order stereochemistry presented limitations in evaluating the application of AlphaFold 3 in modeling protein-glycan interactions.
Putting AlphaFold3 to the Test: Evaluating Structure Prediction Beyond Proteins
In their recent study, researchers from the Department of Biochemistry and Molecular Biology at the University of Georgia modeled a series of glycan and glycan-containing structures to evaluate the capabilities and limitations of the current version of AlphaFold 3. [4] A major challenge arose from the syntax used in glycan modeling, as common input formats such as Simplified Molecular Input Line Entry System (SMILES), Chemical Component Dictionary (CCD) codes, and user-defined CCDs (userCCD) often modeled incorrect stereoisomers. The most accurate results were obtained by employing the Bonded AtomPairs (BAP) syntax to define covalent linkages.
Upon establishing the most effective input format, several glycan–protein complexes were modeled. Among these, the highly branched M9 N-glycan bound to mannosidase MAN1A1 was predicted with relatively high confidence, yielding stereochemically and conformationally plausible models that showed close agreement with available crystallographic data. Additionally, several glycosphigolipids and O-linked glycans were also validated, with promising results.
AlphaFold 3 models with MAN1A1/M9/Ca 2+ Michaelis complex. Extracted from Huang, C.; Kannan, N.; Moremen, K. W. Modeling Glycans with AlphaFold 3: Capabilities, Caveats, and Limitations. Glycobiology 2025, 35 (10), cwaf048
To further assess the capabilities of AlphaFold 3, the complete structure of CD22 (SIGLEC-2) was modeled. This receptor contains four N-glycosylation sites within the IgV domain and seven additional sites distributed across the IgC domains. The resulting model effectively reproduced the receptor’s characteristic behavior, the conformational change induced by the presence of a high-affinity trans-ligand, which disrupts the cis-interactions within the IgV region. Although further empirical validation, this example highlights the potential of AlphaFold 3 to investigate structural hypotheses involving post-translational modifications and dynamic receptor–ligand switching.
N-glycosylated Siglec-2 (CD22) molecules modeled by AlphaFold 3. Extracted from Huang, C.; Kannan, N.; Moremen, K. W. Modeling Glycans with AlphaFold 3: Capabilities, Caveats, and Limitations. Glycobiology 2025, 35 (10), cwaf048
However, all protein crystal structures deposited in the Protein Data Bank (PDB) prior to the cutoff date used for AlphaFold 3 training could potentially correspond to data included in its training set. To prevent replication of previously learned experimental information, recently published structures were employed in this study (AlphaFold 3 is trained with structures published until January 2023).
In each case, the modeled complexes closely reproduced the bound conformation of the glycan ligands observed in the crystal structures, despite these specific complexes not being part of the AlphaFold 3 training dataset. However, the researchers emphasized that glycan modeling in AlphaFold 3 is highly context dependent. They observed multiple instances in which the predicted glycan structures failed to preserve correct stereochemistry or did not accurately replicate the ligand–protein interactions present in the experimental data.
Conclusion
The application of AlphaFold 3 in modeling glycans, both as post-translational modifications and as free ligands, has advanced significantly, and current trends suggest that future updates may enable high-fidelity resolution of glycan structures. However, the present modeling capabilities of AlphaFold 3 still demand substantial expertise in glycochemistry for manual curation, as the existing framework lacks explicit scoring metrics to penalize conformational inaccuracies in glycan predictions. Additionally, the current versiononly offers static snapshots. Glycans are inherently flexible, and their dynamic behavior is crucial to their function. In conclusion, AlphaFold 3 represents a transformative platform for modeling glycan-mediated interactions across proteins, lipids, and carbohydrates; nevertheless, structural validation using orthogonal experimental methods remains essential.
To support researchers in overcoming these challenges, Asparia Glycomics provides tailored services, including custom glycan synthesis, to offer you the best quality of purified saccharides to perform your research and studies. Asparia offers the highest purity in the industry and we fully characterize all our molecules to offer you what you need. Submit a request for your molecule of interest and receive a personalized consultation from our expert team.
References
(1) Varki, A.; Cummings, R. D.; Aebi, M.; Packer, N. H.; Seeberger, P. H.; Esko, J. D.; Stanley, P.; Hart, G.; Darvill, A.; Kinoshita, T.; Prestegard, J. J.; Schnaar, R. L.; Freeze, H. H.; Marth, J. D.; Bertozzi, C. R.; Etzler, M. E.; Frank, M.; Vliegenthart, J. F.; Lütteke, T.; Perez, S.; Bolton, E.; Rudd, P.; Paulson, J.; Kanehisa, M.; Toukach, P.; Aoki-Kinoshita, K. F.; Dell, A.; Narimatsu, H.; York, W.; Taniguchi, N.; Kornfeld, S. Symbol Nomenclature for Graphical Representations of Glycans. Glycobiology 2015, 25 (12), 1323–1324. https://doi.org/10.1093/glycob/cwv091
(2) Fadda, E. Molecular Simulations of Complex Carbohydrates and Glycoconjugates. Curr. Opin. Chem. Biol. 2022, 69, 102175. https://doi.org/10.1016/j.cbpa.2022.102175.
(3) Abramson, J.; Adler, J.; Dunger, J.; Evans, R.; Green, T.; Pritzel, A.; Ronneberger, O.; Willmore, L.; Ballard, A. J.; Bambrick, J.; Bodenstein, S. W.; Evans, D. A.; Hung, C.-C.; O’Neill, M.; Reiman, D.; Tunyasuvunakool, K.; Wu, Z.; Žemgulytė, A.; Arvaniti, E.; Beattie, C.; Bertolli, O.; Bridgland, A.; Cherepanov, A.; Congreve, M.; Cowen-Rivers, A. I.; Cowie, A.; Figurnov, M.; Fuchs, F. B.; Gladman, H.; Jain, R.; Khan, Y. A.; Low, C. M. R.; Perlin, K.; Potapenko, A.; Savy, P.; Singh, S.; Stecula, A.; Thillaisundaram, A.; Tong, C.; Yakneen, S.; Zhong, E. D.; Zielinski, M.; Žídek, A.; Bapst, V.; Kohli, P.; Jaderberg, M.; Hassabis, D.; Jumper, J. M. Accurate Structure Prediction of Biomolecular Interactions with AlphaFold 3. Nature 2024, 630 (8016), 493–500. https://doi.org/10.1038/s41586-024-07487-w.
(4) Huang, C.; Kannan, N.; Moremen, K. W. Modeling Glycans with AlphaFold 3: Capabilities, Caveats, and Limitations. Glycobiology 2025, 35 (10), cwaf048. https://doi.org/10.1093/glycob/cwaf048.