CoChat
Sign Up
PreprintCited by 13Protein Structure and DynamicsMachine Learning in BioinformaticsGeometric and Algebraic Topology

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

Knot or Not? Sequence-Based Identification of Knotted Proteins With Machine Learning

Open access
Šrámková, Denisa, Sikora, Maciej, Uchal, Dawid, Klimentová, Eva, Perlinska, Agata P., Nguyen, Mai Lan, Korpacz, Marta, Malinowska, Roksana, Rubach, Pawel, Šimeček, Petr, Sulkowska, Joanna I.·2023·Protein Science
DOI 10.1101/2023.09.06.556468PMC PMC11184937

Preprint — published as 10.1002/pro.4998 · Paper page

Open this paper in CoChat — free account

Abstract

Knotted proteins, although scarce, are crucial structural components of certain protein families, and their roles remain a topic of intense research. Capitalizing on the vast collection of protein structure predictions offered by AlphaFold, this study computationally examines the entire UniProt database to create a robust dataset of knotted and unknotted proteins. Utilizing this dataset, we develop a machine learning model capable of accurately predicting the presence of knots in protein structures solely from their amino acid sequences, with our best-performing model demonstrating a 98.5% overall accuracy. Unveiling the sequence factors that contribute to knot formation, we discover that proteins predicted to be unknotted from known knotted families are typically non-functional fragments missing a significant portion of the knot core. The study further explores the significance of the substrate binding site in knot formation, particularly within the SPOUT protein family. Our findings spotlight the potential of machine learning in enhancing our understanding of protein topology and propose further investigation into the role of knotted structures across other protein families. TOC Graphic:

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 · 10.1101/2023.09.06.556468

Cites

1 of 1 references verified0 citation errors · 0 not checkable

Statistics

Figures

FIGURE 1 Scheme of the machine learning workflow. The entire approach consists of three main steps: 1. Obtaining data from PDB (with support of Knotprot) and AlphaFold (with support of AlphaKnot) predictions for UniProt; 2. Identifying nontrivial topology in the protein backbone using Topoly; 3. Training a machine learning model to predict nontrivial topology from the protein sequence.
FIGURE 1 Scheme of the machine learning workflow. The entire approach consists of three main steps: 1. Obtaining data from PDB (with support of Knotprot) and AlphaFold (with support of AlphaKnot) predictions for UniProt; 2. Identifying nontrivial topology in the protein backbone using Topoly; 3. Training a machine learning model to predict nontrivial topology from the protein sequence.CC BY · PubMed Central
FIGURE 2 Unknotted proteins aligned with their knotted homologs. The schematic on the left represents the sequential alignment between knotted (green) and unknotted (black) proteins. The region with the knot is marked with a gray area. The unknotted proteins that start or end prematurely in comparison to their knotted counterpart are discarded from further analysis (see the section about SPOUT superfamily for details). We focus on cases in which the unknotted protein can be globally aligned with the knotted homolog (even with a deletion in the knot core) or does not have any knotted homologs (the numbers of these proteins are bolded).
FIGURE 2 Unknotted proteins aligned with their knotted homologs. The schematic on the left represents the sequential alignment between knotted (green) and unknotted (black) proteins. The region with the knot is marked with a gray area. The unknotted proteins that start or end prematurely in comparison to their knotted counterpart are discarded from further analysis (see the section about SPOUT superfamily for details). We focus on cases in which the unknotted protein can be globally aligned with the knotted homolog (even with a deletion in the knot core) or does not have any knotted homologs (the numbers of these proteins are bolded).CC BY · PubMed Central
FIGURE 3 Unknotted membrane proteins from Sodium/calcium exchanger family superposed with their knotted homologs. Upper part: unknotted structure (light orange; UniProtKB ID: A0A5K0X7L3) without the entire knot core, unknotted structure (orange; UniProtKB ID: A0A397EWI1) structure lacking the C‐terminal part of the knot core, and unknotted structure (green; UniProtKB ID: A0A562DKV6) with low‐quality part of the knot core. Lower part: unknotted structure (cyan; UniProtKB ID: A0A453N937) without the N‐terminal and C‐terminal part of the knot core, unknotted structure (magenta; UniProtKB ID: A0A378FTV0) without the N‐terminal part of the knot core, and unknotted structure (light pink; UniProtKB ID: K1RUG8) without the entire knot core. All knotted homologs are shown with gray color (UniProtKB IDs: V4U0P6, A0A397A8U0, E6WVS2, A0A453N959, A0A0E1C8E8, and A0A0L8I7H8, respectively).
FIGURE 3 Unknotted membrane proteins from Sodium/calcium exchanger family superposed with their knotted homologs. Upper part: unknotted structure (light orange; UniProtKB ID: A0A5K0X7L3) without the entire knot core, unknotted structure (orange; UniProtKB ID: A0A397EWI1) structure lacking the C‐terminal part of the knot core, and unknotted structure (green; UniProtKB ID: A0A562DKV6) with low‐quality part of the knot core. Lower part: unknotted structure (cyan; UniProtKB ID: A0A453N937) without the N‐terminal and C‐terminal part of the knot core, unknotted structure (magenta; UniProtKB ID: A0A378FTV0) without the N‐terminal part of the knot core, and unknotted structure (light pink; UniProtKB ID: K1RUG8) without the entire knot core. All knotted homologs are shown with gray color (UniProtKB IDs: V4U0P6, A0A397A8U0, E6WVS2, A0A453N959, A0A0E1C8E8, and A0A0L8I7H8, respectively).CC BY · PubMed Central
FIGURE 4 Overview of protein families with the potential of having both knotted and unknotted topologies. (a) Family DUF4253. A deletion (in red) aligning to the knot loop in the nontrivial homolog (UniProtKB ID: A0A1F0KRL5) creates an unknotted structure (B0C9T3). (b) DUF2254 family. The C‐terminus (in red) instead of passing through the loop (in blue) as in the knotted structure (A0A848U8E4), turns back creating a slipknot—an unknotted structure (D5C1U6). (c) UCH family. Compared with the standard structure with 52 topology (A0A0E0DGQ1), either we observe a different arrangement of the loop committing one crossing (in red), which simplifies the topology to 31 (A0A0E0DGQ2) or the fragment of N‐terminus is missing, which simplifies the topology to an unknot (A0A178ZI05). (d) Sodium/calcium exchanger transmembrane protein family. In the unknotted protein (A0A2G1VBT6) we observe the deletion of the fragment (in red) corresponding to the loop knot in the nontrivial homolog (A0A2S5ZE01).
FIGURE 4 Overview of protein families with the potential of having both knotted and unknotted topologies. (a) Family DUF4253. A deletion (in red) aligning to the knot loop in the nontrivial homolog (UniProtKB ID: A0A1F0KRL5) creates an unknotted structure (B0C9T3). (b) DUF2254 family. The C‐terminus (in red) instead of passing through the loop (in blue) as in the knotted structure (A0A848U8E4), turns back creating a slipknot—an unknotted structure (D5C1U6). (c) UCH family. Compared with the standard structure with 52 topology (A0A0E0DGQ1), either we observe a different arrangement of the loop committing one crossing (in red), which simplifies the topology to 31 (A0A0E0DGQ2) or the fragment of N‐terminus is missing, which simplifies the topology to an unknot (A0A178ZI05). (d) Sodium/calcium exchanger transmembrane protein family. In the unknotted protein (A0A2G1VBT6) we observe the deletion of the fragment (in red) corresponding to the loop knot in the nontrivial homolog (A0A2S5ZE01).CC BY · PubMed Central
FIGURE 5 Unknotted SPOUT proteins (a) superposed with their knotted homologs. Upper part: unknotted structure (magenta; UniProtKB ID: A0A418H8R5) lacking the N‐terminal part of the knot core. Lower part: unknotted structure (blue; UniProtKB ID: A0A090WL68) without the entire knot core, and unknotted structure (orange; UniProtKB ID: A0A356CS99) without the C‐terminal part of the knot core. All knotted homologs are shown with gray color and their knotted regions are marked (UniProtKB IDs: A0A2X1NEC8, A0A090W861, and A0A3D0W667, respectively). Since all of these unknotted proteins lack at least a portion of the knot core, and the knot core is necessary for the SPOUT protein to function, we presume that these proteins are not truly unknotted. (b) Left: an example of a SPOUT protein with a typical 31 knotted structure (UniProtKB ID: R1DT92). Middle: structure with a 31 slipknot topology (UniProtKB ID: A0A424QRT9). The slipknot is formed by a C‐terminal end (colored blue) threaded back through the knot loop. The structure of the knot (colored green) remains typical for the SPOUT superfamily. Right: unknotted structure due to a deletion in the knot core (UniProtKB ID: R1FAN9). The presented structures are models from the AlphaFold database (4th version).
FIGURE 5 Unknotted SPOUT proteins (a) superposed with their knotted homologs. Upper part: unknotted structure (magenta; UniProtKB ID: A0A418H8R5) lacking the N‐terminal part of the knot core. Lower part: unknotted structure (blue; UniProtKB ID: A0A090WL68) without the entire knot core, and unknotted structure (orange; UniProtKB ID: A0A356CS99) without the C‐terminal part of the knot core. All knotted homologs are shown with gray color and their knotted regions are marked (UniProtKB IDs: A0A2X1NEC8, A0A090W861, and A0A3D0W667, respectively). Since all of these unknotted proteins lack at least a portion of the knot core, and the knot core is necessary for the SPOUT protein to function, we presume that these proteins are not truly unknotted. (b) Left: an example of a SPOUT protein with a typical 31 knotted structure (UniProtKB ID: R1DT92). Middle: structure with a 31 slipknot topology (UniProtKB ID: A0A424QRT9). The slipknot is formed by a C‐terminal end (colored blue) threaded back through the knot loop. The structure of the knot (colored green) remains typical for the SPOUT superfamily. Right: unknotted structure due to a deletion in the knot core (UniProtKB ID: R1FAN9). The presented structures are models from the AlphaFold database (4th version).CC BY · PubMed Central
FIGURE 6 Unknotted UCH proteins superposed with their knotted homologs. Upper part: unknotted structure (orange; UniProtKB ID: A0A7S3I8E7) lacking the C‐terminal part of the knot core. Lower part: unknotted structure (cyan; UniProtKB ID: Q06AV7) without the N‐terminal and C‐terminal part of the knot core, unknotted structure (magenta; UniProtKB ID: A0A1A7YJ99) without the N‐terminal part of the knot core, and unknotted structure (green; UniProtKB ID: A0A024WZA) with low‐quality part of the knot core. All knotted homologs are shown with gray color (UniProtKB IDs: A0A078B145, A0A6P6PV05, A0A1A7X9F0, and W7F0M6, respectively).
FIGURE 6 Unknotted UCH proteins superposed with their knotted homologs. Upper part: unknotted structure (orange; UniProtKB ID: A0A7S3I8E7) lacking the C‐terminal part of the knot core. Lower part: unknotted structure (cyan; UniProtKB ID: Q06AV7) without the N‐terminal and C‐terminal part of the knot core, unknotted structure (magenta; UniProtKB ID: A0A1A7YJ99) without the N‐terminal part of the knot core, and unknotted structure (green; UniProtKB ID: A0A024WZA) with low‐quality part of the knot core. All knotted homologs are shown with gray color (UniProtKB IDs: A0A078B145, A0A6P6PV05, A0A1A7X9F0, and W7F0M6, respectively).CC BY · PubMed Central
FIGURE 7 Important regions in SPOUT proteins based on machine learning model. (a) An example of how the model's prediction changes with respect to the patch location based on Nep1 protein (patches of length 10; UniProtKB ID: Q57977). Note the one significant drop inside the knot core. (b) Cumulative positions of the patches (with different lengths: 10, 20, and 40 amino acids) based on the knot core alignment. (c) Logo of the C‐terminal part of the knot core showing the conserved glycine motif. (d) Positions of patches in a TrmD structure (based on the N‐terminal domain of AlphaFold model of UniProtKB ID: B6YQ31). Colored regions represent from the left: knotted region (rainbow) and estimated location of the ligand (magenta); 10 (red), 20 (yellow), and 40 (green) amino acid‐long patches, respectively. Black dots show the position of the conserved glycine residues.
FIGURE 7 Important regions in SPOUT proteins based on machine learning model. (a) An example of how the model's prediction changes with respect to the patch location based on Nep1 protein (patches of length 10; UniProtKB ID: Q57977). Note the one significant drop inside the knot core. (b) Cumulative positions of the patches (with different lengths: 10, 20, and 40 amino acids) based on the knot core alignment. (c) Logo of the C‐terminal part of the knot core showing the conserved glycine motif. (d) Positions of patches in a TrmD structure (based on the N‐terminal domain of AlphaFold model of UniProtKB ID: B6YQ31). Colored regions represent from the left: knotted region (rainbow) and estimated location of the ligand (magenta); 10 (red), 20 (yellow), and 40 (green) amino acid‐long patches, respectively. Black dots show the position of the conserved glycine residues.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

Research works printed but not linked: 39

  1. 10.1021/ja961147m · DOI
  2. 10.1038/nsb0494-213 · DOI
  3. 2019, KnotProt 2.0: a database of proteins with knots and other entangled structures, Nucleic acids research, 47, D367, 10.1093/nar/gky1140 · DOI
  4. 2022, AlphaKnot: server to analyze entanglement in structures predicted by AlphaFold methods, Nucleic Acids Research, 50, W44, 10.1093/nar/gkac388 · DOI
  5. 10.1073/pnas.1205918109 · DOI
  6. 2017, How to fold intricately: using theory and experiments to unravel the properties of knotted proteins, Current Opinion in Structural Biology, 42, 6, 10.1016/j.sbi.2016.10.002 · DOI
  7. 10.1186/1471-2105-8-73 · DOI
  8. 2016, Methyl transfer by substrate signaling from a knotted protein fold, Nature Structural & Molecular Biology, 23, 941, 10.1038/nsmb.3282 · DOI
  9. Zayats V , J. A. J. B. D.-H. S. S. J., Perlinska AP Slipknotted and unknotted monovalent cation-proton antiporters evolved from a common ancestor. PLoS computational biology 2021, 8. · DOI
  10. 2010, Knotted vs. unknotted proteins: evidence of knot-promoting loops, PLoS computational biology, 6, e1000864, 10.1371/journal.pcbi.1000864 · DOI
  11. 10.1016/j.jmb.2007.02.035 · DOI
  12. 10.1073/pnas.0811147106 · DOI
  13. 10.1073/pnas.1007602107 · DOI
  14. Bustamante, A. ; Sotelo-Campos, J. ; Guerra, D. ; Floor, M. ; Wilson, C. ; Bustamante, C. ; Baez, M. The energy cost of polypeptide knot formation and its folding consequences. Nature Communications 2017, 8. · DOI
  15. 2009, Stabilizing effect of knots on proteins, Proceedings of the National Academy of Sciences of the United States of America, 105, 19714
  16. 10.1093/NAR/GKAB1061 · DOI · Paper page
  17. 2022, The integration of AlphaFoldpredicted and crystal structures of human trans-3-hydroxy-l-proline dehydratase reveals a regulatory catalytic mechanism, Computational and Structural Biotechnology Journal, 20, 3874, 10.1016/j.csbj.2022.07.027 · DOI
  18. da Silva, F. B. ; Lewandowska, I. ; Kluza, A. ; Niewieczerzal, S. ; Augustyniak, R. ; Sulkowska, J. I. First crystal structure of double knotted protein TrmD-Tm1570 – inside from degradation perspective. bioRxiv 2023, · DOI
  19. Brems, M. A. ; Runkel, R. ; Yeates, T. O. ; Virnau, P. AlphaFold predicts the most complex protein knot and composite protein knots. Protein Science 2022, 31. · DOI
  20. Perlinska, A. P. ; Niemyska, W. H. ; Gren, B. A. ; Bukowicki, M. ; Nowakowski, S. ; Rubach, P. ; Sulkowska, J. I. scpAlphaFold/scp predicts novel human proteins with knots. Protein Science 2023, 32. · DOI
  21. 10.1111/j.1742-4658.2009.06990.x · DOI
  22. 2006, Intricate knots in proteins: Function and evolution, PLoS computational biology, 2, e122, 10.1371/journal.pcbi.0020122 · DOI
  23. Strassler, S. E. ; Bowles, I. E. ; Dey, D. ; Jackman, J. E. ; Conn, G. L. Tied up in knots: Untangling substrate recognition by the SPOUT methyltransferases. Journal of Biological Chemistry 2022, 298. · DOI
  24. 2020, Topoly: Python package to analyze topology of polymers, Briefings in Bioinformatics, 22, bbaa196
  25. Freyd, P. ; Yetter, D. ; Hoste, J. ; Lickorish, W. ; Millett, K. ; Ocneanu, A. A new polynomial invariant of knots and links. Bull. Amer. Math. Soc. 1985, 12. · DOI
  26. 2019, Proteins’ knotty problems, Journal of Molecular Biology, 431, 244, 10.1016/j.jmb.2018.10.012 · DOI
  27. Sikora, M. ; Flapan, E. ; Wong, H. ; Rubach, P. ; Garstka, W. ; Niewieczerzal, S. ; Rawdon, E. J. ; Sulkowska, J. I. Proteins containing 6-crossing knot types and their folding pathways. bioRxiv 2023, 2023–06. · DOI
  28. 2022, ProtTrans: Toward Understanding the Language of Life Through Self-Supervised Learning, IEEE Transactions on Pattern Analysis and Machine Intelligence, 44, 7112, 10.1109/TPAMI.2021.3095381 · DOI · Paper page
  29. 10.1038/nrg2776 · DOI · Paper page
  30. 2021, Slipknotted and unknotted monovalent cation-proton antiporters evolved from a common ancestor, PLoS Computational Biology, 17, e1009502, 10.1371/journal.pcbi.1009502 · DOI
  31. Shi, D. ; Allewell, N. M. ; Tuchman, M. From genome to structure and back again: A family portrait of the transcarbamylases. International Journal of Molecular Sciences 2015, 16. · DOI
  32. 2020, Restriction of S-adenosylmethionine conformational freedom by knotted protein binding sites, PLoS computational biology, 16, e1007904, 10.1371/journal.pcbi.1007904 · DOI
  33. Lundberg, S. M. ; Lee, S.-I. A unified approach to interpreting model predictions. Advances in neural information processing systems 2017, 30.
  34. 2022, InterPro in 2022, Nucleic Acids Research, 51, D418
  35. 10.1093/nar/gky1140 · DOI
  36. 2014, KnotProt: a database of proteins with knots and slipknots, Nucleic Acids Research, 43, D306
  37. 10.1093/bioinformatics/bts565 · DOI · Paper page
  38. Paszke, A. et al. Advances in Neural Information Processing Systems 32 ; Curran Associates, Inc., 2019; pp 8024–8035.
  39. Cik, I. ; Rasamoelina, A. D. ; Mach, M. ; Sincak, P. Explaining Deep Neural Network using Layer-wise Relevance Propagation and Integrated Gradients. 2021. · DOI

Metadata

Authors
Šrámková, D., Sikora, M., Uchal, D., Klimentová, E., Perlinska, A. P., Nguyen, M. L., Korpacz, M., Malinowska, R., Rubach, P., Šimeček, P., & Sulkowska, J. I.
Affiliations
  • Masaryk University (CZ)
  • University of Warsaw (PL)
  • Central European Institute of Technology – Masaryk University (CZ)
Corresponding author
  • Šimeček, Petr
  • Sulkowska, Joanna I.
ORCID
Venue
Protein Science
Publisher
openRxiv
Published
2023-09-07
Volume
33
Issue
7
Pages
e4998-e4998
ISSN
0961-8368 (ISSN-L)
DOI
10.1101/2023.09.06.556468
PMC
PMC11184937
OpenAlex
W4399776685
Cited by
13
References
1
Citations by year
2026: 3 · 2025: 5 · 2024: 5
FWCI
1.77
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Protein Science (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nd)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • ReIn UW (University of Warsaw) (submittedVersion, cc-by-nc-nd)
Funders
  • Grantová Agentura České Republiky (GF23-04260L)
  • Narodowe Centrum Nauki (#2021/43/I/NZ1/03341, UMO-2018/31/B/NZ1/04016)
  • Masarykova Univerzita (LM2018140, 90254, CZ LM2018140)
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
  • Physical Sciences › Mathematics › Geometry and Topology › Geometric and Algebraic Topology
Keywords
knotted proteins, sequence-based prediction, protein topology, AlphaFold, UniProt
MeSH terms
Machine Learning, Amino Acid Sequence, Models, Molecular, Protein Conformation, Proteins, Databases, Protein

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