CoChat
Sign Up
Cited by 127Protein Structure and DynamicsMachine Learning in BioinformaticsEnzyme Structure and Function

The DOI matches its Crossref record; 72 of the 72 references it cites were found in a registry or an index.

Ins and outs of AlphaFold2 transmembrane protein structure predictions

Open access
Hegedűs, Tamás, Geisler, Markus, Lukács, Gergely László, Farkas, Bianka·2022·Cellular and Molecular Life Sciences
DOI 10.1007/s00018-021-04112-1PMC PMC8761152
Open this paper in CoChat — free accountOpen PDF

Abstract

Transmembrane (TM) proteins are major drug targets, but their structure determination, a prerequisite for rational drug design, remains challenging. Recently, the DeepMind's AlphaFold2 machine learning method greatly expanded the structural coverage of sequences with high accuracy. Since the employed algorithm did not take specific properties of TM proteins into account, the reliability of the generated TM structures should be assessed. Therefore, we quantitatively investigated the quality of structures at genome scales, at the level of ABC protein superfamily folds and for specific membrane proteins (e.g. dimer modeling and stability in molecular dynamics simulations). We tested template-free structure prediction with a challenging TM CASP14 target and several TM protein structures published after AlphaFold2 training. Our results suggest that AlphaFold2 performs well in the case of TM proteins and its neural network is not overfitted. We conclude that cautious applications of AlphaFold2 structural models will advance TM protein-associated studies at an unexpected level.

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 · 10.1007/s00018-021-04112-1

Cites

72 of 72 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Quantitative analysis of human AF TM structures. (a) Mean pLDDT scores were calculated for human transmembrane (TMEM), soluble (SOLU), TM regions of TM proteins, and non-TM regions of the same proteins. The fraction of structures in reliability ranges, used in the human proteome AlphaFold2 paper [5], are shown. (b) The hydrophobic thickness was calculated for human TM proteins as the distance between the center of geometry of Cα atoms of side1 and side2 of transmembrane helices. TM helices of AF2-structures were selected based on CCTOP predictions. The hydrophobic thickness of experimental structures was collected from PDBTM. The inset demonstrates how the distance calculation can be effected by a topology in the case of incorrectly built AF2 structures (purple; correct structure: green; s1 and s2: side1 and side2). (c) The hydrophobic thickness of each protein and the corresponding CCTOP reliability scores are shown. (d) The hydrophobic thickness of each protein and the corresponding pLDDT scores were plotted
Fig. 1 Quantitative analysis of human AF TM structures. (a) Mean pLDDT scores were calculated for human transmembrane (TMEM), soluble (SOLU), TM regions of TM proteins, and non-TM regions of the same proteins. The fraction of structures in reliability ranges, used in the human proteome AlphaFold2 paper [5], are shown. (b) The hydrophobic thickness was calculated for human TM proteins as the distance between the center of geometry of Cα atoms of side1 and side2 of transmembrane helices. TM helices of AF2-structures were selected based on CCTOP predictions. The hydrophobic thickness of experimental structures was collected from PDBTM. The inset demonstrates how the distance calculation can be effected by a topology in the case of incorrectly built AF2 structures (purple; correct structure: green; s1 and s2: side1 and side2). (c) The hydrophobic thickness of each protein and the corresponding CCTOP reliability scores are shown. (d) The hydrophobic thickness of each protein and the corresponding pLDDT scores were plottedCC BY · PubMed Central
Fig. 2 All AF2-predicted ABC structures exhibit valid ABC TM folds. (a) The best TM-score for every experimental ABC TM structure and ABC reference fold comparisons are shown in the boxplot, grouped into ABC fold families, sorted by the total number of included transmembrane chains. Numbers indicate the sum of chains in experimental structures matching an ABC reference fold. Only MlaE fold family does not contain experimental structures released before 2018-04-30. (b) The same plot was generated for ABC protein structures from the 21 proteomes predicted by AF2. The number of matched structures within a fold family is indicated in parenthesis. A TM-score above 0.5 indicates that the compared structure and the reference fold exhibits the same architecture
Fig. 2 All AF2-predicted ABC structures exhibit valid ABC TM folds. (a) The best TM-score for every experimental ABC TM structure and ABC reference fold comparisons are shown in the boxplot, grouped into ABC fold families, sorted by the total number of included transmembrane chains. Numbers indicate the sum of chains in experimental structures matching an ABC reference fold. Only MlaE fold family does not contain experimental structures released before 2018-04-30. (b) The same plot was generated for ABC protein structures from the 21 proteomes predicted by AF2. The number of matched structures within a fold family is indicated in parenthesis. A TM-score above 0.5 indicates that the compared structure and the reference fold exhibits the same architectureCC BY · PubMed Central
Fig. 3 Blind transmembrane fold predictions. The AF2-predicted structures (blue to red: N- to C-termini) of the challenging CASP14 TM target, LmrP (a), the MlaE (b), and the EMC6 (c) exhibit perfect alignments with their experimental structures (gray) 6t1z, 7ch0, and 6ww7, respectively. None of these experimental structures were published before 2018-04-30. Only structures with a sequence homologous to LmrP were in the AF2 training set
Fig. 3 Blind transmembrane fold predictions. The AF2-predicted structures (blue to red: N- to C-termini) of the challenging CASP14 TM target, LmrP (a), the MlaE (b), and the EMC6 (c) exhibit perfect alignments with their experimental structures (gray) 6t1z, 7ch0, and 6ww7, respectively. None of these experimental structures were published before 2018-04-30. Only structures with a sequence homologous to LmrP were in the AF2 training setCC BY · PubMed Central
Fig. 4 AF2 predicts two conformations of a new transmembrane fold. (a) The top AF2-prediction of the novel MprF TM fold (blue to red: N- to C-termini) aligned to the experimental structure 7duw (gray). (b) pLDDT and TM-score values, calculated for every structural model from six runs, were plotted. Numbers (1–5) indicate the corresponding AF2 models. Red points were the top ranked hits from a given run. (c) Structural alignment of the prediction with the best pLDDT score (blue to red) and experimental structure (gray). 7duw and any other structure homologous to MprF were not included in the AF2 training set
Fig. 4 AF2 predicts two conformations of a new transmembrane fold. (a) The top AF2-prediction of the novel MprF TM fold (blue to red: N- to C-termini) aligned to the experimental structure 7duw (gray). (b) pLDDT and TM-score values, calculated for every structural model from six runs, were plotted. Numbers (1–5) indicate the corresponding AF2 models. Red points were the top ranked hits from a given run. (c) Structural alignment of the prediction with the best pLDDT score (blue to red) and experimental structure (gray). 7duw and any other structure homologous to MprF were not included in the AF2 training setCC BY · PubMed Central
Fig. 5 AF2 predictions and ABC structure-associated issues. (a) ABCG2 and ABCG8 NBD β1 strand sequence alignments generated by structural alignment of 6hco (ABCG2) and 5do7 (ABCG5/ABCG8), by ClustalW with manual adjustment of ABCG2 and ABCG8 sequences based on ABCG2 structures, and by structural alignment of ABCG2 and AF2-predicted ABCG8 NBDs. Structure: AF2 ABCG8 NBD, blue: β1 strand, red: the segment corresponding to the β1 strand in the registry shifted 5do7 NBD, cyan: gating loop or regulatory insertion. (b) Structural alignment of 7jr7 (gray) and AF2-predicted (blue) ABCG5/ABCG8 TM domains (top view). Non-conserved loops with low-quality predictions are red. (c) Aligned homology (orange: TMD1, red: TMD2) and AF2 (blue: TMD1, cyan: TMD2) models of AtABCG36. Blue and orange spheres label F589 and F592 in TM2 facing the substrate binding pocket. (d) The magnified view of AtABCG36 TM1 and TM2 indicates that the alignments are not shifted but that spatial localization and side chain packing differ. (e) TM2 in the homology model unwinds in MD simulations. (f) zfCFTR TM8 is kinked in PDBID:5w81 (red) along with other structures and it is straight in both MRP1-based model (orange) and AF2-predicted structure (blue). The helices are extracted from a full TM domain alignment for visualization. (g) Surface representation of zfCFTR (PDBID:5w81). Red: TM8, green: TMD1, cyan: TMD2, pale green: NBD1, pale cyan: NBD2, black spheres: CAVER spheres indicating channel opening towards the extracellular space and the extracellular boundary of the lipid bilayer. (h) Surface representation of zfCFTR with MRP1-modelled, straight TM8. No lateral opening to the extracellular membrane leaflet can be observed
Fig. 5 AF2 predictions and ABC structure-associated issues. (a) ABCG2 and ABCG8 NBD β1 strand sequence alignments generated by structural alignment of 6hco (ABCG2) and 5do7 (ABCG5/ABCG8), by ClustalW with manual adjustment of ABCG2 and ABCG8 sequences based on ABCG2 structures, and by structural alignment of ABCG2 and AF2-predicted ABCG8 NBDs. Structure: AF2 ABCG8 NBD, blue: β1 strand, red: the segment corresponding to the β1 strand in the registry shifted 5do7 NBD, cyan: gating loop or regulatory insertion. (b) Structural alignment of 7jr7 (gray) and AF2-predicted (blue) ABCG5/ABCG8 TM domains (top view). Non-conserved loops with low-quality predictions are red. (c) Aligned homology (orange: TMD1, red: TMD2) and AF2 (blue: TMD1, cyan: TMD2) models of AtABCG36. Blue and orange spheres label F589 and F592 in TM2 facing the substrate binding pocket. (d) The magnified view of AtABCG36 TM1 and TM2 indicates that the alignments are not shifted but that spatial localization and side chain packing differ. (e) TM2 in the homology model unwinds in MD simulations. (f) zfCFTR TM8 is kinked in PDBID:5w81 (red) along with other structures and it is straight in both MRP1-based model (orange) and AF2-predicted structure (blue). The helices are extracted from a full TM domain alignment for visualization. (g) Surface representation of zfCFTR (PDBID:5w81). Red: TM8, green: TMD1, cyan: TMD2, pale green: NBD1, pale cyan: NBD2, black spheres: CAVER spheres indicating channel opening towards the extracellular space and the extracellular boundary of the lipid bilayer. (h) Surface representation of zfCFTR with MRP1-modelled, straight TM8. No lateral opening to the extracellular membrane leaflet can be observedCC 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

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

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

Show all 69 references

Research works printed but not linked: 1

  1. McCoy AJ, Sammito MD, Read RJ (2021) Possible implications of AlphaFold2 for crystallographic phasing by molecular replacement. BioRxiv. https://doi.org/10.1101/2021.05.18.444614 · DOI

Cited by

Sorted by citations

See all 127

Related

Metadata

Authors
Hegedűs, T., Geisler, M., Lukács, G. L., & Farkas, B.
Affiliations
  • Semmelweis University (HU)
  • University of Fribourg (CH)
  • McGill University (CA)
Corresponding author
Hegedűs, Tamás
ORCID
Venue
Cellular and Molecular Life Sciences
Publisher
Springer Science and Business Media LLC
Published
2022-01
Volume
79
Issue
1
Pages
73-73
ISSN
  • 1420-682X (ISSN-L)
  • 1420-9071
DOI
10.1007/s00018-021-04112-1
PMC
PMC8761152
OpenAlex
W4205172550
Cited by
127
References
69
Citations by year
2026: 15 · 2025: 19 · 2024: 23 · 2023: 41 · 2022: 28
FWCI
9.06
Type
journal article
Language
en
Open access
Licence
Locations
  • Cellular and Molecular Life Sciences (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
  • Nemzeti Kutatási Fejlesztési és Innovációs Hivatal (K127961)
  • Cystic Fibrosis Foundation (HEGEDU20I0)
  • Cystic Fibrosis Canada (LUKACS20G0)
  • Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (310030_197563)
  • Canadian Institutes of Health Research
  • Canada Foundation for Innovation
  • Canada Research Chairs
  • Semmelweis University
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
AlphaFold2, transmembrane proteins, protein structure prediction, prediction accuracy assessment, ABC transporters, membrane protein dimerization, CASP14, overfitting, structural genomics
MeSH terms
Machine Learning, Protein Domains, Algorithms, Computer Simulation, Humans, Lipids, Membrane Proteins, Protein Conformation, Software, Reproducibility of Results, Genome, Protein Structure, Secondary, Protein Folding, Computational Biology, Proteome, Proteomics, Molecular Dynamics Simulation

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