CoChat
Sign Up
PreprintCited by 3Protein Structure and DynamicsMachine Learning in BioinformaticsEnzyme Structure and Function

The DOI matches its Crossref record; 2 of the 2 references it cites were found in a registry or an index.

CCfrag: Scanning folding potential of coiled-coil fragments with AlphaFold

Open access
Martinez-Goikoetxea, Mikel·2024·Bioinformatics Advances
DOI 10.1101/2024.05.24.595610PMC PMC11676326

Preprint — published as 10.1093/bioadv/vbae195

Open this paper in CoChat — free accountOpen PDF

Abstract

Structured abstract: Motivation: Coiled coils are a widespread structural motif consisting of multiple α-helices that wind around a central axis to bury their hydrophobic core. Although their backbone can be uniquely described by the Crick parametric equations, these have little practical application in structural prediction, given that most coiled coils in nature feature non-canonical repeats that locally distort their geometry. While AlphaFold has emerged as an effective coiled-coil modeling tool, capable of accurately predicting changes in periodicity and core geometry along coiled-coil stalks, it is not without limitations. These include the generation of spuriously bent models and the inability to effectively model globally non-canonical coiled coils. In an effort to overcome these limitations, we investigated whether dividing full-length sequences into fragments would result in better models. Results: We developed CCfrag to leverage AlphaFold for the piece-wise modeling of coiled coils. The user can create a specification, defined by window size, length of overlap, and oligomerization state, and the program produces the files necessary to run structural predictions with AlphaFold. Then, the structural models and their scores are integrated into a rich per-residue representation defined by sequence-or structure-based features, which can be visualized or employed for further analysis. Our results suggest that removing coiled-coil sequences from their native context can in some case improve the prediction confidence and avoids bent models with spurious contacts. In this paper, we present various use cases of CCfrag, and propose that fragment-based prediction is useful for understanding the properties of long, fibrous coiled coils, by showing local features not seen in full-length models. Availability and Implementation: The program is implemented as a Python module. The code and its documentation are available at https://github.com/Mikel-MG/CCfrag . Contact: mikel.martinez@tuebingen.mpg.de

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 · 10.1101/2024.05.24.595610

Cites

2 of 2 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1. Schematic representation of the CCfrag pipeline. Modeling long coiled-coil domains with AlphaFold generally yields suboptimal models; (A, B) A full-length model of Homo sapiens EEA1 is shown, colored by pLDDT (red – worst to blue – best). By dividing the full-length sequence into fragments (C), the resulting models are predicted with higher confidence (D), and can be analysed for local properties not seen in the full-length model (E).
Figure 1. Schematic representation of the CCfrag pipeline. Modeling long coiled-coil domains with AlphaFold generally yields suboptimal models; (A, B) A full-length model of Homo sapiens EEA1 is shown, colored by pLDDT (red – worst to blue – best). By dividing the full-length sequence into fragments (C), the resulting models are predicted with higher confidence (D), and can be analysed for local properties not seen in the full-length model (E).CC BY · PubMed Central
Figure 2. Comparison between full-length and piece-wise modeling of EEA1 from Homo sapiens. (A) pLDDT plot of the full-length dimer model. (B) Graphical summary of the CCfrag representation of the protein. It is modeled as a dimer in four different specifications with window sizes of 20, 30, 50, and 70 residues and an overlap of half the corresponding window size. Plots for various features are shown (pLDDT, PAE, bundle orientation, KIH interactions) for each specification (color-coded as in the legend in the left of the fourth panel). The features are averaged for each residue (since the overlap is half the window size, each residue is covered by two models, resulting in two values per-residue). In the bottom panel, coiled-coil prediction probabilities for COILS (window size = 21) and DeepCoil 2 are shown. (C) Boxplot representations for the pLDDT scores across different specifications, including the full-length model. A cartoon render of the full-length model is shown in Fig. 1B.
Figure 2. Comparison between full-length and piece-wise modeling of EEA1 from Homo sapiens. (A) pLDDT plot of the full-length dimer model. (B) Graphical summary of the CCfrag representation of the protein. It is modeled as a dimer in four different specifications with window sizes of 20, 30, 50, and 70 residues and an overlap of half the corresponding window size. Plots for various features are shown (pLDDT, PAE, bundle orientation, KIH interactions) for each specification (color-coded as in the legend in the left of the fourth panel). The features are averaged for each residue (since the overlap is half the window size, each residue is covered by two models, resulting in two values per-residue). In the bottom panel, coiled-coil prediction probabilities for COILS (window size = 21) and DeepCoil 2 are shown. (C) Boxplot representations for the pLDDT scores across different specifications, including the full-length model. A cartoon render of the full-length model is shown in Fig. 1B.CC BY · PubMed Central
Figure 3. Graphical summary of the CCfrag representation of the spike protein of human SARS coronavirus (UniProt P59594). The spike protein is involved in the fusion between the viral and cellular membranes, and although it is not a fibrous protein, it features two functionally relevant coiled-coil domains, known as HR1 and HR2 (for heptad repeat). (A) The protein is modeled as a dimer, a trimer, and a tetramer, each in 50-residue windows with a 25-residue overlap. There are three regions which are more confidently predicted than their surrounding context (high pLDDT, low PAE), two of which correspond to HR1 and HR2. (B) (Left) Fragment 725–775 of the protein is predicted as an interlocking array of coiled coils. (Center) Fragment 925–975 is predicted as a parallel trimer, nearly identical to the experimentally solved structure (1ZVB, in grey; RMSD: 0.4 Angstroms). (Right) Fragment 1150–1200 is predicted as an antiparallel tetramer, although with a register different of that of the experimentally solved structure (PDB 1ZV7, in grey). The predicted structures are colored by pLDDT (red – worst to blue – best).
Figure 3. Graphical summary of the CCfrag representation of the spike protein of human SARS coronavirus (UniProt P59594). The spike protein is involved in the fusion between the viral and cellular membranes, and although it is not a fibrous protein, it features two functionally relevant coiled-coil domains, known as HR1 and HR2 (for heptad repeat). (A) The protein is modeled as a dimer, a trimer, and a tetramer, each in 50-residue windows with a 25-residue overlap. There are three regions which are more confidently predicted than their surrounding context (high pLDDT, low PAE), two of which correspond to HR1 and HR2. (B) (Left) Fragment 725–775 of the protein is predicted as an interlocking array of coiled coils. (Center) Fragment 925–975 is predicted as a parallel trimer, nearly identical to the experimentally solved structure (1ZVB, in grey; RMSD: 0.4 Angstroms). (Right) Fragment 1150–1200 is predicted as an antiparallel tetramer, although with a register different of that of the experimentally solved structure (PDB 1ZV7, in grey). The predicted structures are colored by pLDDT (red – worst to blue – best).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: 17

  1. 10.1107/S0365110X53001964 · DOI · Paper page
  2. 10.1107/S0365110X53001952 · DOI · Paper page
  3. 2006, Structures and Polymorphic Interactions of Two Heptad-Repeat Regions of the SARS Virus S2 Protein, Structure (London, England: 1993), 14, 889, 10.1016/j.str.2006.03.007 · DOI
  4. Evans R , O’Neill M , Pritzel A et al. Protein Complex Prediction with AlphaFold-Multimer. Bioinformatics, 2021. · DOI · Paper page
  5. 10.1126/science.8248779 · DOI · Paper page
  6. 10.1093/protein/gzm071 · DOI
  7. 2021, Socket2: a program for locating, visualizing and analyzing coiled-coil interfaces in protein structures, Bioinformatics, 37, 4575, 10.1093/bioinformatics/btab631 · DOI · Paper page
  8. Lin Z , Akin H , Rao R et al. Evolutionary-Scale Prediction of Atomic Level Protein Structure with a Language Model. Synthetic Biology, 2022. · DOI · Paper page
  9. 10.1093/bioinformatics/bty1062 · DOI
  10. 10.1126/science.252.5009.1162 · DOI · Paper page
  11. 10.1016/j.tibs.2016.10.007 · DOI
  12. 2017, The Structure and Topology of α-Helical Coiled Coils, Fibrous Proteins: Structures and Mechanisms, 82, 95
  13. Madaj R , Martinez-Goikoetxea M , Kaminski K et al. Applicability of AlphaFold2 in the modelling of coiled-coil domains. 2024:2024.03.07.583852. · DOI
  14. 2023, New protein families with hendecad coiled coils in the proteome of life, Journal of Structural Biology, 215, 108007, 10.1016/j.jsb.2023.108007 · DOI
  15. 10.1038/nature19326 · DOI
  16. 10.1016/J.CELL.2024.01.022 · DOI · Paper page
  17. 10.1002/pro.4846 · DOI

Metadata

Authors
Martinez-Goikoetxea, M.
ORCID
Martinez-Goikoetxea, Mikel
Venue
Bioinformatics Advances
Publisher
openRxiv
Published
2024-05-30
Volume
5
Issue
1
Pages
vbae195-vbae195
ISSN
2635-0041 (ISSN-L)
DOI
10.1101/2024.05.24.595610
PMC
PMC11676326
OpenAlex
W4405223295
Cited by
3
References
2
Citations by year
2026: 1 · 2025: 2
FWCI
0.41
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Bioinformatics Advances (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
Max-Planck-Gesellschaft
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
coiled coil, AlphaFold, oligomeric state, fibrous proteins, protein structure prediction

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