CoChat
Sign Up
Cited by 16Protein Structure and DynamicsMachine Learning in BioinformaticsEnzyme Structure and Function

The DOI matches its Crossref record; 6 of the 6 references it cites were found in a registry or an index.

Balancing speed and precision in protein folding: a comparison of AlphaFold2, ESMFold, and OmegaFold

Open access
Hýsková, Anna, Maršálková, Eva, Šimeček, Petr·2026·Frontiers in Genetics
DOI 10.3389/fgene.2025.1715037PMC PMC12844563

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

The rapid development of protein structure prediction tools has created a need for systematic performance comparisons to guide method selection for different applications, particularly given the trade-offs between computational speed and prediction accuracy. We benchmarked AlphaFold2, ESMFold, and OmegaFold using 1,337 protein chains deposited in the Protein Data Bank between July 2022 and July 2024, ensuring no overlap with training data, and evaluated predictions using Root Mean Square Deviation (RMSD), Template Modeling score (TM-score), Global Distance Test–Total Score (GDT-TS) and predicted Local Distance Difference Test (pLDDT) metrics. AlphaFold2 achieved the highest median TM-score (0.96), highest median GDT-TS (94%), and lowest median RMSD (1.30 Å), outperforming ESMFold (TM-score 0.95, GDT-TS 90%, RMSD 1.74 Å) and OmegaFold (TM-score 0.93, GDT-TS 89%, RMSD 1.98 Å), with all tools showing reduced accuracy for proteins lacking family annotations, leucine-rich repeats, and NMR-determined structures, while alignment-free methods unexpectedly excelled at de novo designed proteins. The performance differences between methods were negligible for many proteins, suggesting that faster alignment-free predictors (10–30 times faster) can be sufficient for numerous applications; we developed LightGBM classifiers using ProtBert embeddings and confidence scores that accurately predict when AlphaFold2’s computational investment is warranted, providing practitioners with actionable guidance for selecting between speed and precision in structural pipelines.

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 · 10.3389/fgene.2025.1715037

Cites

6 of 6 references verified0 citation errors · 0 not checkable

Statistics

Figures

FIGURE 1 Examples of structure predictions from AlphaFold2 (red), ESMFold (blue) and OmegaFold (yellow) aligned with corresponding experimentally determined structures (green). (a) An example of a poorly predicted structure (8P4Y:A) by AlphaFold2. (b) Structure of protein 8PTF:A showing varying prediction quality across tools.
FIGURE 1 Examples of structure predictions from AlphaFold2 (red), ESMFold (blue) and OmegaFold (yellow) aligned with corresponding experimentally determined structures (green). (a) An example of a poorly predicted structure (8P4Y:A) by AlphaFold2. (b) Structure of protein 8PTF:A showing varying prediction quality across tools.CC BY · PubMed Central
FIGURE 2 Dependency of average pLDDT, TM-score, RMSD, and GDT-TS on the type of protein chain being predicted. The differences between groups were tested by Kruskal–Wallis test, post hoc comparisons were done using Dunn’s method with a Bonferroni correction for multiple tests. Different letters above boxplots indicate statistically significant differences among label groups within each prediction tool (compact letter display; groups sharing a letter are not significantly different). Sample points with RMSD greater than 30 Å are omitted from the visualization for better clarity.
FIGURE 2 Dependency of average pLDDT, TM-score, RMSD, and GDT-TS on the type of protein chain being predicted. The differences between groups were tested by Kruskal–Wallis test, post hoc comparisons were done using Dunn’s method with a Bonferroni correction for multiple tests. Different letters above boxplots indicate statistically significant differences among label groups within each prediction tool (compact letter display; groups sharing a letter are not significantly different). Sample points with RMSD greater than 30 Å are omitted from the visualization for better clarity.CC BY · PubMed Central
FIGURE 3 Comparison of the overlap of poorly predicted protein chains. (a) Venn diagrams show the overlap of poorly predicted chains among the three structure prediction tools (AlphaFold2, ESMFold, and OmegaFold) for each evaluation metric: average pLDDT <70, TM-score <0.5, RMSD >9Å, and GDT-TS <50%. (b) UpSet plots visualize the intra-tool failure patterns by quantifying how many predictions fail each individual metric and every combination of metrics.
FIGURE 3 Comparison of the overlap of poorly predicted protein chains. (a) Venn diagrams show the overlap of poorly predicted chains among the three structure prediction tools (AlphaFold2, ESMFold, and OmegaFold) for each evaluation metric: average pLDDT <70, TM-score <0.5, RMSD >9Å, and GDT-TS <50%. (b) UpSet plots visualize the intra-tool failure patterns by quantifying how many predictions fail each individual metric and every combination of metrics.CC BY · PubMed Central
FIGURE 4 SHAP analysis of LightGBM models predicting TM-score. SHAP summary plots for LightGBM regressors trained separately for AlphaFold2, ESMFold, and OmegaFold using ProtBert sequence embeddings, sequence length, experimental acquisition method, and model-specific confidence estimates (pLDDT) as input features. Features are ordered by mean absolute SHAP value, indicating their overall influence on predicted TM-score. Each point represents an individual protein chain, colored by feature contribution.
FIGURE 4 SHAP analysis of LightGBM models predicting TM-score. SHAP summary plots for LightGBM regressors trained separately for AlphaFold2, ESMFold, and OmegaFold using ProtBert sequence embeddings, sequence length, experimental acquisition method, and model-specific confidence estimates (pLDDT) as input features. Features are ordered by mean absolute SHAP value, indicating their overall influence on predicted TM-score. Each point represents an individual protein chain, colored by feature contribution.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

Show all 6 references

Research works printed but not linked: 32

  1. Alberts, 2017, Molecular biology of the cell, 10.1201/9781315735368 · DOI
  2. Bittrich, 2023, RCSB protein data bank: efficient searching and simultaneous access to one million computed structure models alongside the PDB structures enabled by architectural advances, J. Mol. Biol., 435, 167994, 10.1016/j.jmb.2023.167994 · DOI
  3. Brandes, 2022, Proteinbert: a universal deep-learning model of protein sequence and function, Bioinformatics, 38, 2102, 10.1093/bioinformatics/btac020 · DOI · Paper page
  4. Callaway, 2024, Ai protein-prediction tool alphafold3 is now more open, Nature, 635, 531, 10.1038/d41586-024-03708-4 · DOI
  5. 2024, Chai-1: decoding the molecular interactions of life, bioRxiv, 10.1101/2024.10.10.615955 · DOI · Paper page
  6. Cock, 2009, Biopython: freely available Python tools for computational molecular biology and bioinformatics, Bioinformatics, 25, 1422, 10.1093/bioinformatics/btp163 · DOI · Paper page
  7. Cooper, 2000, The cell: a molecular approach
  8. Elliott, 2025, Abcfold: easier running and comparison of alphafold 3, boltz-1 and chai-1, Bioinforma. Adv., 5, vbaf153, 10.1093/bioadv/vbaf153 · DOI · Paper page
  9. Hu, 2021, NMR-based methods for protein analysis, Anal. Chem., 93, 1866, 10.1021/acs.analchem.0c03830 · DOI
  10. Huang, 2023, Protein structure prediction: challenges, advances, and the shift of research paradigms, Genomics Proteomics Bioinform., 21, 913, 10.1016/j.gpb.2022.11.014 · DOI
  11. Kalogeropoulos, 2024, A comparative study of protein structure prediction tools for challenging targets: snake venom toxins, Toxicon, 238, 107559, 10.1016/j.toxicon.2023.107559 · DOI
  12. Ke, 2017, Lightgbm: a highly efficient gradient boosting decision tree, Adv. Neural Inform. Process. Syst., 30, 10.5555/3294996.3295074 · DOI
  13. Kovalevskiy, 2024, AlphaFold two years on: validation and impact, Proc. Natl. Acad. Sci., 121, e2315002121, 10.1073/pnas.2315002121 · DOI
  14. Kufareva, 2012, Methods of protein structure comparison, Methods Mol. Biol., 857, 231, 10.1007/978-1-61779-588-6_10 · DOI
  15. Manfredi, 2024, Alpha&ESMhFolds: a web server for comparing AlphaFold2 and ESMFold models of the human reference proteome, J. Mol. Biol., 436, 168593, 10.1016/j.jmb.2024.168593 · DOI
  16. Manfredi, 2025, Evaluation of the structural models of the human reference proteome: Alphafold2 versus esmfold, Curr. Res. Struct. Biol., 9, 100167, 10.1016/j.crstbi.2025.100167 · DOI
  17. Milne, 2013, Cryo-electron microscopy: a primer for the non-microscopist, FEBS J., 280, 28, 10.1111/febs.12078 · DOI · Paper page
  18. Moult, 1995, A large-scale experiment to assess protein structure prediction methods, Proteins Struct. Funct. Bioinforma., 23, ii, 10.1002/prot.340230303 · DOI · Paper page
  19. Moussad, 2023, The transformative power of transformers in protein structure prediction, Proc. Natl. Acad. Sci., 120, e2303499120, 10.1073/pnas.2303499120 · DOI
  20. 2020, R: a language and environment for statistical computing
  21. Robin, 2021, Continuous automated model EvaluatiOn (CAMEO)—Perspectives on the future of fully automated evaluation of structure prediction methods, Proteins Struct. Funct. Bioinforma., 89, 1977, 10.1002/prot.26213 · DOI
  22. Rose, 2021, RCSB protein data bank: architectural advances towards integrated searching and efficient access to macromolecular structure data from the PDB archive, J. Mol. Biol., 433, 166704, 10.1016/j.jmb.2020.11.003 · DOI
  23. Selkoe, 2003, Folding proteins in fatal ways, Nature, 426, 900, 10.1038/nature02264 · DOI
  24. Škrinjar, 2025, Have protein-ligand co-folding methods moved beyond memorisation?, BioRxiv, 10.1101/2025.02.03.636309 · DOI
  25. Smyth, 2000, X ray crystallography, Mol. Pathol., 53, 8, 10.1136/mp.53.1.8 · DOI
  26. Valdés-Tresanco, 2023, Structural modeling of nanobodies: a benchmark of state-of-the-art artificial intelligence programs, Molecules, 28, 3991, 10.3390/molecules28103991 · DOI
  27. Wohlwend, 2024, Boltz-1: democratizing biomolecular interaction modeling, bioRxiv, 10.1101/2024.11.19.624167 · DOI
  28. Wu, 2022, High-resolution de novo structure prediction from primary sequence, Bioinformatics, 10.1101/2022.07.21.500999 · DOI · Paper page
  29. Xu, 2025, Foldbench: an all-atom benchmark for biomolecular structure prediction, bioRxiv, 10.1101/2025.05.22.655600 · DOI
  30. Zemla, 2003, Lga: a method for finding 3d similarities in protein structures, Nucleic Acids Res., 31, 3370, 10.1093/nar/gkg571 · DOI · Paper page
  31. Zhang, 2004, Scoring function for automated assessment of protein structure template quality, Proteins Struct. Funct. Bioinforma., 57, 702, 10.1002/prot.20264 · DOI · Paper page
  32. Zhou, 2025, Benchmarking alphafold3-like methods for protein-peptide complex prediction, bioRxiv, 10.1101/2025.03.09.642277 · DOI

Metadata

Authors
Hýsková, A., Maršálková, E., & Šimeček, P.
Affiliations
  • Central European Institute of Technology – Masaryk University (CZ)
  • Faculty of Science, Central European Institute of Technology (CEITEC), Masaryk University
Corresponding author
Šimeček, Petr
ORCID
Šimeček, Petr
Venue
Frontiers in Genetics
Publisher
Frontiers Media SA
Published
2026-01-14
Volume
16
Pages
1715037-1715037
ISSN
1664-8021 (ISSN-L)
DOI
10.3389/fgene.2025.1715037
PMC
PMC12844563
OpenAlex
W7123648101
Cited by
16
References
6
Citations by year
2026: 14
FWCI
28.93
Type
journal article
Language
en
Open access
Licence
Locations
  • Frontiers in Genetics (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Veřejné služby Informačního systému (Masarykiana Brunensis Universitas) (submittedVersion, cc-by)
Funders
Grantová Agentura České Republiky
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
protein structure prediction, AlphaFold2, ESMFold, benchmarking, prediction accuracy, computational speed, leucine-rich repeats, NMR structure, method selection

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Balancing speed and precision in protein folding: a comparison of AlphaFold2, ESMFold, and OmegaFold. https://app.cochat.ai/papers/10.3389/fgene.2025.1715037
The paper
Hýsková, A., Maršálková, E., & Šimeček, P. (2026). Balancing speed and precision in protein folding: a comparison of AlphaFold2, ESMFold, and OmegaFold. Frontiers in Genetics. https://doi.org/10.3389/fgene.2025.1715037