rewire.it
Configuration

MULAN-ESM2 S

MULAN-ESM2 S augments a pretrained ESM-2 sequence encoder with a structure adapter.

SourcesMULAN: multimodal protein language model for sequence and structure encoding · 2 Methods/2.1 MULAN architecture/2.1.1 Structural information (paragraph 1); 3 Discussion/3.2 Ablation study/3.2.1 The importance of the Structure Adapter (paragraph 1)

1 evaluation · 1 metric row

How it worksEvaluated procedure (conceptual)
Evaluated procedure (conceptual)1. Protein amino-acid sequences and angle-based structural descriptors. Then: 2. MULAN-ESM2 S. Then: 3. Structure-aware representations for downstream protein tasksEvaluated procedure (conceptual)1. Protein amino-acid sequences and angle-based structural descriptors. Then: 2. MULAN-ESM2 S. Then: 3. Structure-aware representations for downstream protein tasksEvaluated procedure (conceptual)1. Protein amino-acid sequences and angle-based structural descriptors. Then: 2. MULAN-ESM2 S. Then: 3. Structure-aware representations for downstream protein tasks

Conceptual input–method–output guide. Check the procedure text and linked evaluation for fitted components, additional inputs and exact settings.

SourcesMULAN: multimodal protein language model for sequence and structure encoding · 2 Methods/2.2 Training procedure and structure masking (paragraph 1); 2 Methods/2.1 MULAN architecture/2.1.2 The Structure Adapter (paragraph 1)

At a glance

Model type

Protein sequence transformer; this record is the paper-specific evaluated configuration.

Sourcesfacebookresearch/esm README.md · README.md model description

limited source coverage · Automated source review, 2026-09-16. All specifications and missing details

Evaluations and results

Release 2026-09-17-d277315f7d76 · 1 evaluation · 1 metric row. Different protocols are not a single leaderboard.

Results grouped by the exact reported evaluation
Metric and findingCoverage and uncertaintyEvidence
MULAN-ESM2 S: human protein-protein interaction prediction

MULAN sequence-structure model based on ESM2 8M

Author-reported evaluation · Evaluation metadata: needs review

0.717 AUC

Unit: fraction · Direction: unknown

Uncertainty: not reported in legacy extract

Scored: Not reported · Eligible: Not reported

source checkedMULAN: multimodal protein language model for sequence and structure encoding · Table 2, MULAN-ESM2 S row, HumanPPI AUC column

Source checking is not independent reproduction.

How it works

How the evaluated method works

A parameter-efficient adapter encodes angle-based structural information, which is fused with the pretrained sequence representation and trained jointly.

SourcesMULAN: multimodal protein language model for sequence and structure encoding · 2 Methods/2.2 Training procedure and structure masking (paragraph 1); 2 Methods/2.1 MULAN architecture/2.1.2 The Structure Adapter (paragraph 1)
Underlying method and version boundaries

ESM-2 is a transformer protein language-model family. The official repository exposes residue embeddings, sequence-level pooling and models at several sizes; the study configuration determines which of these is evaluated.

Sourcesfacebookresearch/esm README.md · README.md; introduction, model description, pretrained-model and usage sections at pinned revision
What was evaluated

The linked evaluation record identifies MULAN-ESM2 S: human protein-protein interaction prediction. Its dataset, split, adaptation and evidence origin remain attached to the reported results.

SourcesMULAN: multimodal protein language model for sequence and structure encoding · The named evaluation’s methods and comparison table; exact preserved evaluation IDs: evaluation-b2-mulan-2025

Strengths and limitations

Profile review details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Stable record: reported-model-a29203c09857ef

Specifications

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.

Inputs, outputs and configuration
PropertyDescription and evidence
Model typeProtein sequence transformer; this record is the paper-specific evaluated configuration.
Sourcesfacebookresearch/esm README.md · README.md model description
Architecture / procedureA parameter-efficient adapter encodes angle-based structural information, which is fused with the pretrained sequence representation and trained jointly.
SourcesMULAN: multimodal protein language model for sequence and structure encoding · 2 Methods/2.2 Training procedure and structure masking (paragraph 1); 2 Methods/2.1 MULAN architecture/2.1.2 The Structure Adapter (paragraph 1)
Biological inputsProtein amino-acid sequences and angle-based structural descriptors
SourcesMULAN: multimodal protein language model for sequence and structure encoding · 2 Methods/2.1 MULAN architecture/2.1.1 Structural information (paragraph 2); 3 Discussion/3.4 Related work/3.4.2 Structure-informed models (paragraph 1)
OutputsStructure-aware representations for downstream protein tasks
SourcesMULAN: multimodal protein language model for sequence and structure encoding · 3 Discussion/3.1 Results (paragraph 1); 4 Conclusion (paragraph 1)
ParametersMULAN-ESM2 S has 9M parameters, including its ESM2-8M backbone and structure adapter. The 37M and 652M models are separate variants.
SourcesMULAN: multimodal protein language model for sequence and structure encoding · 3 Discussion/3.2 Ablation study/3.2.1 The importance of the Structure Adapter (paragraph 1); 2 Methods/2.1 MULAN architecture/2.1.1 Structural information (paragraph 1)
Known versions / configurationsmall ESM2 backbone
SourcesMULAN: multimodal protein language model for sequence and structure encoding · Table vbaf117-T4 (paragraph 1); Table vbaf117-T2 (paragraph 1)
Training data / fitting501,348 Swiss-Prot proteins with AlphaFold2 structures after filtering sequences shorter than 30 residues; 5,000 proteins are selected for validation. The pretrained backbone is further trained with joint sequence/angle masking.
SourcesMULAN: multimodal protein language model for sequence and structure encoding · 2 Methods/2.3 Training details (paragraph 1); 2 Methods/2.2 Training procedure and structure masking (paragraph 1)
Context limitsA maximum input/context length for this exact evaluated configuration is not established by the inspected sources. · Not reported in inspected sources
Sources (2)MULAN: multimodal protein language model for sequence and structure encoding; facebookresearch/esm README.md · 2 Methods/2.1 MULAN architecture/2.1.1 Structural information; 2 Methods/2.1 MULAN architecture/2.1.2 The Structure Adapter; 2 Methods/2.2 Training procedure and structure masking; 2 Methods/2.3 Training details; 2 Methods/2.4 Downstream tasks; 2 Methods/2.5 Downstream task evaluation; 3 Discussion/3.2 Ablation study/3.2.4 Architecture ablation; inspected for explicit maximum input length (dataset lengths and family-wide limits are not substituted); README.md at pinned repository revision
AccessOfficial upstream implementation and usage documentation: https://github.com/facebookresearch/esm/blob/2b369911bb5b4b0dda914521b9475cad1656b2ac/README.md. This pinned documentation revision is not automatically the evaluated weight revision.
Sourcesfacebookresearch/esm README.md · README.md; installation, model download and usage instructions
Code licenceMIT (upstream repository code at the cited revision; this does not establish every dependency or historical checkpoint licence).
Sourcesfacebookresearch/esm LICENSE · LICENSE; complete licence text
Weights licenceThe inspected model-access documentation does not explicitly identify terms for this exact evaluated checkpoint or fitted head; repository code terms are shown separately. · Not reported in inspected sources
Sourcesfacebookresearch/esm README.md · README.md; checkpoint/access documentation and licence scope

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.

21 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-09-17-d277315f7d76
Property and statementOriginal source and locationReview and provenance
Diagram caption

Conceptual input–method–output guide. Check the procedure text and linked evaluation for fitted components, additional inputs and exact settings.

Individual claims
MULAN: multimodal protein language model for sequence and structure encoding

Original source ↗

2 Methods/2.2 Training procedure and structure masking (paragraph 1); 2 Methods/2.1 MULAN architecture/2.1.2 The Structure Adapter (paragraph 1)

Version: journal full text in PMC
Retrieved: 2026-09-16T10:38:57.558217+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Field: attributes.profile.diagram.caption

Source artifact SHA-256: 771a9a26ebda6f49ea266540e8dd6e6de0cbaef724de818ca6124a5f9c50d350

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Diagram steps

["Protein amino-acid sequences and angle-based structural descriptors","MULAN-ESM2 S","Structure-aware representations for downstream protein tasks"]

Individual claims
MULAN: multimodal protein language model for sequence and structure encoding

Original source ↗

2 Methods/2.2 Training procedure and structure masking (paragraph 1); 2 Methods/2.1 MULAN architecture/2.1.2 The Structure Adapter (paragraph 1)

Version: journal full text in PMC
Retrieved: 2026-09-16T10:38:57.558217+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Field: attributes.profile.diagram.steps

Source artifact SHA-256: 771a9a26ebda6f49ea266540e8dd6e6de0cbaef724de818ca6124a5f9c50d350

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Diagram title

Evaluated procedure (conceptual)

Individual claims
MULAN: multimodal protein language model for sequence and structure encoding

Original source ↗

2 Methods/2.2 Training procedure and structure masking (paragraph 1); 2 Methods/2.1 MULAN architecture/2.1.2 The Structure Adapter (paragraph 1)

Version: journal full text in PMC
Retrieved: 2026-09-16T10:38:57.558217+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Field: attributes.profile.diagram.title

Source artifact SHA-256: 771a9a26ebda6f49ea266540e8dd6e6de0cbaef724de818ca6124a5f9c50d350

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Model type

Protein sequence transformer; this record is the paper-specific evaluated configuration.

Individual claims
facebookresearch/esm README.md

Original source ↗

README.md model description

Version: 2b369911bb5b4b0dda914521b9475cad1656b2ac
Retrieved: 2026-09-16T20:00:00.816433+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Field: attributes.profile.facts.0.value

Source artifact SHA-256: 8b273c21a322fc9473d1b68d0dd40c8166ab2f89e4a190aa26ca87251b97cba9

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Architecture / procedure

A parameter-efficient adapter encodes angle-based structural information, which is fused with the pretrained sequence representation and trained jointly.

Individual claims
MULAN: multimodal protein language model for sequence and structure encoding

Original source ↗

2 Methods/2.2 Training procedure and structure masking (paragraph 1); 2 Methods/2.1 MULAN architecture/2.1.2 The Structure Adapter (paragraph 1)

Version: journal full text in PMC
Retrieved: 2026-09-16T10:38:57.558217+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Field: attributes.profile.facts.1.value

Source artifact SHA-256: 771a9a26ebda6f49ea266540e8dd6e6de0cbaef724de818ca6124a5f9c50d350

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Weights licence

The inspected model-access documentation does not explicitly identify terms for this exact evaluated checkpoint or fitted head; repository code terms are shown separately.

Individual claims
facebookresearch/esm README.md

Original source ↗

README.md; checkpoint/access documentation and licence scope

Version: 2b369911bb5b4b0dda914521b9475cad1656b2ac
Retrieved: 2026-09-16T20:00:00.816433+00:00

unreported

automated source review · 2026-09-16

Audit details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Field: attributes.profile.facts.10.value

Source artifact SHA-256: 8b273c21a322fc9473d1b68d0dd40c8166ab2f89e4a190aa26ca87251b97cba9

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Biological inputs

Protein amino-acid sequences and angle-based structural descriptors

Individual claims
MULAN: multimodal protein language model for sequence and structure encoding

Original source ↗

2 Methods/2.1 MULAN architecture/2.1.1 Structural information (paragraph 2); 3 Discussion/3.4 Related work/3.4.2 Structure-informed models (paragraph 1)

Version: journal full text in PMC
Retrieved: 2026-09-16T10:38:57.558217+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Field: attributes.profile.facts.2.value

Source artifact SHA-256: 771a9a26ebda6f49ea266540e8dd6e6de0cbaef724de818ca6124a5f9c50d350

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Outputs

Structure-aware representations for downstream protein tasks

Individual claims
MULAN: multimodal protein language model for sequence and structure encoding

Original source ↗

3 Discussion/3.1 Results (paragraph 1); 4 Conclusion (paragraph 1)

Version: journal full text in PMC
Retrieved: 2026-09-16T10:38:57.558217+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Field: attributes.profile.facts.3.value

Source artifact SHA-256: 771a9a26ebda6f49ea266540e8dd6e6de0cbaef724de818ca6124a5f9c50d350

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Parameters

MULAN-ESM2 S has 9M parameters, including its ESM2-8M backbone and structure adapter. The 37M and 652M models are separate variants.

Individual claims
MULAN: multimodal protein language model for sequence and structure encoding

Original source ↗

3 Discussion/3.2 Ablation study/3.2.1 The importance of the Structure Adapter (paragraph 1); 2 Methods/2.1 MULAN architecture/2.1.1 Structural information (paragraph 1)

Version: journal full text in PMC
Retrieved: 2026-09-16T10:38:57.558217+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Field: attributes.profile.facts.4.value

Source artifact SHA-256: 771a9a26ebda6f49ea266540e8dd6e6de0cbaef724de818ca6124a5f9c50d350

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Known versions / configuration

small ESM2 backbone

Individual claims
MULAN: multimodal protein language model for sequence and structure encoding

Original source ↗

Table vbaf117-T4 (paragraph 1); Table vbaf117-T2 (paragraph 1)

Version: journal full text in PMC
Retrieved: 2026-09-16T10:38:57.558217+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary full text and the available official implementation/model documentation were inspected. Explanatory claims are source-backed; unresolved exact-configuration metadata is labelled explicitly. This is automated review, not a human review or independent benchmark reproduction.

Field: attributes.profile.facts.5.value

Source artifact SHA-256: 771a9a26ebda6f49ea266540e8dd6e6de0cbaef724de818ca6124a5f9c50d350

Hash scope: Hash scope not separately documented; inspect source record

Inspected artifact

Sources and history

Release 2026-09-17-d277315f7d76 · Record review: needs review

3 source records and release historyDownload this release
Technical metadata and extraction receipts

Stable ID: reported-model-a29203c09857ef

areas
proteins-complexes
entity level
method
version
small ESM2 backbone
reported name
MULAN-ESM2 S
historical missing metadata
checkpoint revision: not_reported_in_legacy_extract; training data: not_reported_in_legacy_extract; licence: not_reported_in_legacy_extract
metadata review scope
historical_missing_metadata preserves the original discovery state. Current descriptive evidence and missingness are recorded in profile.facts; numerical-result review is separate.
legacy kinds
model
entity classification
review date: 2026-09-17; rationale: This source-scoped entry preserves the method/configuration actually named in an evaluation. It is neither a global family identity nor proof of an immutable checkpoint; the linked evaluation retains adaptation, fitting and scoring details.; source ids: mulan-2025; evidence-reported-base-esm-readme-md; source locator: 2 Methods/2.2 Training procedure and structure masking (paragraph 1); 2 Methods/2.1 MULAN architecture/2.1.2 The Structure Adapter (paragraph 1) | README.md model description | 2 Methods/2.1 MULAN architecture/2.1.1 Structural information (paragraph 1); 3 Discussion/3.2 Ablation study/3.2.1 The importance of the Structure Adapter (paragraph 1); ambiguities: Configuration means the source-labelled evaluated identity. It does not establish missing checkpoint hashes, default settings or equivalence to same-named records in other papers.
Related records

Suggest a correction