Model type
Sequence-to-structure protein prediction pipeline
Configuration
ESMFold
ESMFold predicts protein structures directly from amino-acid sequence using ESM-2 representations.
Sources (3)
facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_configHow it worksESMFold workflow
Conceptual summary of the documented data flow; optional inputs and configured downstream stages must be reported for a reproducible evaluation.
Sources (3)
facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_configOverview
Inputs
Protein amino-acid sequence; the ESMFold interface also accepts chains separated by a colon.
Outputs
Predicted PDB structure and confidence values.
Access
Official project documentation and implementation: https://github.com/facebookresearch/esm
Sources (3)
facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_configlimited source coverage · Automated source review, 2026-09-16. All specifications and missing details
Evaluations and results
14 evaluations · 14 results. Different protocols are not a single leaderboard.
Filter evaluations
Applied filters: All linked evaluations
| Tested configuration | Protocol and dataset | Finding | Evidence and details |
|---|---|---|---|
| Configuration: ESMFold | Task: ProteinBench FOLD-ACCURACY-GDT-TS-MEAN: Accuracy GDT-TS (mean) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 0.826 gdt-ts_mean score · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-ACCURACY-GDT-TS-MEAN: Accuracy GDT-TS (mean) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Accuracy GDT-TS ↑) |
| Configuration: ESMFold | Task: ProteinBench FOLD-ACCURACY-GDT-TS-MEDIAN: Accuracy GDT-TS (median) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 0.881 gdt-ts_median score · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-ACCURACY-GDT-TS-MEDIAN: Accuracy GDT-TS (median) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Accuracy GDT-TS ↑) |
| Configuration: ESMFold | Task: ProteinBench FOLD-ACCURACY-LDDT-MEAN: Accuracy lDDT (mean) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 0.87 lddt_mean score · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-ACCURACY-LDDT-MEAN: Accuracy lDDT (mean) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Accuracy lDDT ↑) |
| Configuration: ESMFold | Task: ProteinBench FOLD-ACCURACY-LDDT-MEDIAN: Accuracy lDDT (median) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 0.907 lddt_median score · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-ACCURACY-LDDT-MEDIAN: Accuracy lDDT (median) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Accuracy lDDT ↑) |
| Configuration: ESMFold | Task: ProteinBench FOLD-ACCURACY-RMSD-MEAN: Accuracy RMSD (mean) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 3.98 rmsd_mean score · lower Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-ACCURACY-RMSD-MEAN: Accuracy RMSD (mean) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Accuracy RMSD ↓) |
| Configuration: ESMFold | Task: ProteinBench FOLD-ACCURACY-RMSD-MEDIAN: Accuracy RMSD (median) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 2.1 rmsd_median score · lower Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-ACCURACY-RMSD-MEDIAN: Accuracy RMSD (median) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Accuracy RMSD ↓) |
| Configuration: ESMFold | Task: ProteinBench FOLD-ACCURACY-TM-SCORE-MEAN: Accuracy TM-score (mean) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 0.847 tm-score_mean score · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-ACCURACY-TM-SCORE-MEAN: Accuracy TM-score (mean) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Accuracy TM-score ↑) |
| Configuration: ESMFold | Task: ProteinBench FOLD-ACCURACY-TM-SCORE-MEDIAN: Accuracy TM-score (median) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 0.929 tm-score_median score · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-ACCURACY-TM-SCORE-MEDIAN: Accuracy TM-score (median) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Accuracy TM-score ↑) |
| Configuration: ESMFold | Task: ProteinBench FOLD-QUALITY-CA-BREAK-MEAN: Quality CA break (%) (mean) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 0% ca-break_mean percent · lower Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-QUALITY-CA-BREAK-MEAN: Quality CA break (%) (mean) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Quality CA break (%) ↓) |
| Configuration: ESMFold | Task: ProteinBench FOLD-QUALITY-CA-BREAK-MEDIAN: Quality CA break (%) (median) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 0% ca-break_median percent · lower Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-QUALITY-CA-BREAK-MEDIAN: Quality CA break (%) (median) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Quality CA break (%) ↓) |
| Configuration: ESMFold | Task: ProteinBench FOLD-QUALITY-CA-CLASH-MEAN: Quality CA clash (%) (mean) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 0.3% ca-clash_mean percent · lower Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-QUALITY-CA-CLASH-MEAN: Quality CA clash (%) (mean) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Quality CA clash (%) ↓) |
| Configuration: ESMFold | Task: ProteinBench FOLD-QUALITY-CA-CLASH-MEDIAN: Quality CA clash (%) (median) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 0% ca-clash_median percent · lower Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-QUALITY-CA-CLASH-MEDIAN: Quality CA clash (%) (median) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Quality CA clash (%) ↓) |
| Configuration: ESMFold | Task: ProteinBench FOLD-QUALITY-PEPBOND-BREAK-MEAN: Quality PepBond break (%) (mean) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 4.7% pepbond-break_mean percent · lower Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-QUALITY-PEPBOND-BREAK-MEAN: Quality PepBond break (%) (mean) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Quality PepBond break (%) ↓) |
| Configuration: ESMFold | Task: ProteinBench FOLD-QUALITY-PEPBOND-BREAK-MEDIAN: Quality PepBond break (%) (median) Dataset subset: CAMEO2022, 183 proteins (ProteinBench split) | 3.4% pepbond-break_median percent · lower Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESMFold on ProteinBench FOLD-QUALITY-PEPBOND-BREAK-MEDIAN: Quality PepBond break (%) (median) Structure prediction on CAMEO2022. Each cell prints the mean and the median over 183 proteins. Aggregation: Not reported proteinbench primary benchmark evidence · Table 7, row(ESMFold), column(Quality PepBond break (%) ↓) |
Source checking is not independent reproduction. Release 2026-09-29-06401fd5b220.
Use this model
How it works, versions and access
Related profile: ESMFold. This page retains the exact record and its evaluation context.
This configuration
Protein model evaluated by the ProteinBench authors under their harness.
- record
- ESMFold
- configuration
- Not reported
- entity type
- Configuration
How it works
How it works
ESMFold predicts protein structures directly from amino-acid sequence using ESM-2 representations. ESM-2 sequence representations feed a folding trunk and structure module. The checked v1 configuration has 48 trunk blocks, eight structure-module blocks and up to four recycles. The documented inputs are protein amino-acid sequence; the ESMFold interface also accepts chains separated by a colon. The output consists of predicted PDB structure and confidence values.
Sources (3)
facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_configVersions and reproducibility
esmfold_v0 and esmfold_v1; v1 is the repository recommendation. Inference length is constrained by memory; the repository documents chunking and CPU offload. Backbone position settings do not alone establish the full folding pipeline limit.
Sources (3)
facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_configStrengths, limitations and unresolved questions
Strengths and limitations
Strengths and considerations
- Supports sequence-only structure prediction without an MSA search.
Sources (3)
facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config
Limitations and conditions
- ESMFold v0 and v1 are different releases. The repository discourages using structure-module-only ablation models as the standard predictor.
Sources (3)
facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config
Profile review details
Inspected pinned official documentation, relevant implementation files and named primary-paper sections. Claims are limited to those artifacts. Remaining field extraction and identity conflicts are explicit; no new performance claims, model runs or human review are implied.
Stable record: discovery-model-esmfoldSpecifications
Inputs, training, access and other details
Explanatory profile: limited source coverage · Automated source review, 2026-09-16. Review applies to the cited claims; unresolved fields are listed below. Numerical results retain their own review status.
| Property | Description and evidence |
|---|---|
| Model type | Sequence-to-structure protein prediction pipelineSources (3)facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config |
| Architecture | ESM-2 sequence representations feed a folding trunk and structure module. The checked v1 configuration has 48 trunk blocks, eight structure-module blocks and up to four recycles.Sources (3)facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config |
| Inputs | Protein amino-acid sequence; the ESMFold interface also accepts chains separated by a colon.Sources (3)facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config |
| Outputs | Predicted PDB structure and confidence values.Sources (3)facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config |
| Parameters | The released v1 configuration identifies an ESM-2 3B backbone plus a folding trunk and structure module. The 3B figure is not the total size of the complete predictor.Sources (3)facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config |
| Known versions | esmfold_v0 and esmfold_v1; v1 is the repository recommendation.Sources (3)facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config |
| Training data | PDB and UniRef50 are listed for ESMFold. Full structural training-cutoff verification remains outstanding.Sources (3)facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config |
| Training cutoff | PDB and UniRef50 are identified in the official model table; the inspected ESMFold-v1 card and configuration do not supply a shared latest-data date for both components. · Not reported in inspected sourcesSources (3)facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config |
| Context limits | Inference length is constrained by memory; the repository documents chunking and CPU offload. Backbone position settings do not alone establish the full folding pipeline limit.Sources (3)facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config |
| Weights licence | MIT declared by the official facebook/esmfold_v1 model card.Sources (3)facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config |
| Access | Official project documentation and implementation: https://github.com/facebookresearch/esmSources (3)facebookresearch/esm: README.md; facebook/esmfold_v1: README.md; facebook/esmfold_v1: config.json · ESM README: ESMFold Structure Prediction; official facebook/esmfold_v1 model card and config.json esmfold_config |
| Code licence | MITSourcesfacebookresearch/esm: LICENSE · LICENSE: licence text |
Evidence
Source checking verifies the cited claim or transcription. It does not establish independent reproduction.
Evidence table
Inspect claims, sources and review details
Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.
One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.
2 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|---|---|
| Relationship: family catalog-model-esmfold Individual claims | proteinbench primary benchmark evidence Appendix B.2.3 Model implementations, ESMFold; Table7 CAMEO2022 Version: 2409.06744v1 | source checked automated source review · 2026-09-23 Audit detailsSource review establishes this relationship only. Exact evaluated configurations and original numerical review status remain unchanged. Primary paper explicitly states esm.pretrained.esmfold_v1() and one deterministic structure per protein. This is an exact v1 identity statement, not inferred from the current default API. Field: Claim: model-evaluation-identity-4fbbf97bfa5a3a627b52 Source artifact SHA-256: Hash scope: Exact retrieved primary paper artifact bytes. |
| Relationship: family discovery-model-esmfold Individual claims | proteinbench primary benchmark evidence Tables 7; task methods and corresponding named row Version: 2409.06744v1 | source checked automated source review · 2026-09-23 Audit detailsSource review establishes this relationship only. Exact evaluated configurations and original numerical review status remain unchanged. Paper evaluates this named model under its ProteinBench harness. Association is to the family, not an assertion of checkpoint equivalence or cross-task comparability. Field: Claim: model-evaluation-identity-2c12c5c6d204ab99ce83 Source artifact SHA-256: Hash scope: Exact retrieved primary paper artifact bytes. |
Sources and history
View linked audit checks and correction history
Release 2026-09-29-06401fd5b220 · Record review: source checked
1 source records and release history
- proteinbench primary benchmark evidence · Original source · 2409.06744v1
Technical metadata and extraction receipts
Stable ID: proteinbench-method-esmfold
- areas
- proteins-complexes
- source locator
- Table 7, row(ESMFold)
- missing metadata
- checkpoint revision: unreported; parameters: unextracted
Related records
- family: ESMFold
- family: ESMFold
- subject: ESMFold: family ESMFold
- subject: ESMFold: family ESMFold
- model: ESMFold on ProteinBench FOLD-ACCURACY-GDT-TS-MEAN: Accuracy GDT-TS (mean)
- model: ESMFold on ProteinBench FOLD-ACCURACY-GDT-TS-MEDIAN: Accuracy GDT-TS (median)
- model: ESMFold on ProteinBench FOLD-ACCURACY-LDDT-MEAN: Accuracy lDDT (mean)
- model: ESMFold on ProteinBench FOLD-ACCURACY-LDDT-MEDIAN: Accuracy lDDT (median)
- model: ESMFold on ProteinBench FOLD-ACCURACY-RMSD-MEAN: Accuracy RMSD (mean)
- model: ESMFold on ProteinBench FOLD-ACCURACY-RMSD-MEDIAN: Accuracy RMSD (median)
- model: ESMFold on ProteinBench FOLD-ACCURACY-TM-SCORE-MEAN: Accuracy TM-score (mean)
- model: ESMFold on ProteinBench FOLD-ACCURACY-TM-SCORE-MEDIAN: Accuracy TM-score (median)
- model: ESMFold on ProteinBench FOLD-QUALITY-CA-BREAK-MEAN: Quality CA break (%) (mean)
- model: ESMFold on ProteinBench FOLD-QUALITY-CA-BREAK-MEDIAN: Quality CA break (%) (median)
- model: ESMFold on ProteinBench FOLD-QUALITY-CA-CLASH-MEAN: Quality CA clash (%) (mean)
- model: ESMFold on ProteinBench FOLD-QUALITY-CA-CLASH-MEDIAN: Quality CA clash (%) (median)
- model: ESMFold on ProteinBench FOLD-QUALITY-PEPBOND-BREAK-MEAN: Quality PepBond break (%) (mean)
- model: ESMFold on ProteinBench FOLD-QUALITY-PEPBOND-BREAK-MEDIAN: Quality PepBond break (%) (median)