rewire.itbenchmarks
← Back to results
Benchmark

GlycanML

GlycanML evaluates glycan learning across multiple classification and interaction tasks.

Sources (5)GlycanML/GlycanML official source; GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml · Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction

323 evaluations · 323 results

Overview

Datasets

Glycan taxonomy, immunogenicity, glycosylation-type and protein–glycan interaction datasets.

Metrics

Checked single-task configurations use accuracy/MCC for taxonomy and glycosylation type, AUROC/AUPRC for immunogenicity, and MAE/RMSE/Spearman for protein–glycan interaction regression.

Allowed inputs

Glycan sequence or graph representations and, for interaction tasks, paired protein data.

Sources (5)GlycanML/GlycanML official source; GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml · Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction
Evaluation procedure diagram
How it worksEvaluation procedure
Evaluation procedure1. Allowed inputs: Glycan sequence or graph representations and, for interaction tasks, paired protein data.. Then: 2. Splits: Taxonomy, immunogenicity and glycosylation tasks use motif-cluster partitions. Protein–glycan interaction uses MMseqs2 protein clusters with threshold 0.5. Both allocate clusters 8:1:1; the two notions of held-out data are different.. Then: 3. Metrics: Checked single-task configurations use accuracy/MCC for taxonomy and glycosylation type, AUROC/AUPRC for immunogenicity, and MAE/RMSE/Spearman for protein–glycan interaction regression.Evaluation procedure1. Allowed inputs: Glycan sequence or graph representations and, for interaction tasks, paired protein data.. Then: 2. Splits: Taxonomy, immunogenicity and glycosylation tasks use motif-cluster partitions. Protein–glycan interaction uses MMseqs2 protein clusters with threshold 0.5. Both allocate clusters 8:1:1; the two notions of held-out data are different.. Then: 3. Metrics: Checked single-task configurations use accuracy/MCC for taxonomy and glycosylation type, AUROC/AUPRC for immunogenicity, and MAE/RMSE/Spearman for protein–glycan interaction regression.Evaluation procedure1. Allowed inputs: Glycan sequence or graph representations and, for interaction tasks, paired protein data.. Then: 2. Splits: Taxonomy, immunogenicity and glycosylation tasks use motif-cluster partitions. Protein–glycan interaction uses MMseqs2 protein clusters with threshold 0.5. Both allocate clusters 8:1:1; the two notions of held-out data are different.. Then: 3. Metrics: Checked single-task configurations use accuracy/MCC for taxonomy and glycosylation type, AUROC/AUPRC for immunogenicity, and MAE/RMSE/Spearman for protein–glycan interaction regression.

Conceptual procedure. Task variants and protocol versions retain their separate scoring conditions.

Sources (6)GlycanML/GlycanML official source; GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml; glycanml primary benchmark evidence · Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; Sections 3.1–3.4

Source reviewed · Automated source review, 2026-09-16. All specifications and missing details

Results

Each comparison retains its reviewed evaluation scope, dataset and metric. Results are shown without a pooled ranking.

GlycanML Domain · Table 3, p. 8

accuracy (percent) · Higher values are better.

SugarBase taxonomy · Domain (GlycanML taxonomy prediction) · SugarBase taxonomy · Domain

Evidence origin: Author-reported evaluation.

GlycanML: A Multi-Task and Multi-Structure Benchmark for Glycan Machine Learning · Table 3, p. 8, Domain
  • Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.
  • Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.
  • Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.
Comparison details and limitations

Author-reported single-task models on the same held-out task; not a cross-paper ranking. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.

Automated source review: 2026-09-17. Numerical source review does not establish independent reproduction.

Dots show point estimates. Whiskers show only explicitly defined uncertainty (standard deviation, standard error or a labelled interval); their definitions remain in Table. Unresolved uncertainty is not plotted. Differences do not establish statistical significance.

Showing 10 of 10 matching rows.

Tested configuration
0255075100
Reported score
  1. RGCN94.78(0.11)
  2. GIN94.41(0.17)
  3. GCN94.38(0.51)
  4. GAT94.27(0.41)
  5. CompGCN93.94(0.41)
  6. MPNN93.83(0.17)
  7. Shallow CNN93.76(0.66)
  8. ResNet93.29(1.46)
  9. LSTM92.78(1.16)
  10. Transformer91.98(0.27)

Methods and evaluation design

Procedure, tasks and evaluated configurations

How it works

Evaluation methodology

GlycanML evaluates taxonomy, immunogenicity, glycosylation type and protein–glycan interaction. Glycans are encoded as IUPAC sequences or graphs. Structural motif clusters define the first three task partitions, whereas interaction prediction holds out protein sequence clusters and predicts a transformed fluorescence binding signal.

Sourcesglycanml primary benchmark evidence · Sections 3.1–3.4

Evaluation design

Benchmarks bring together tasks and protocols. A task describes the biological question; a protocol defines a particular test.

Tasks

Protocols

These source-backed links do not make different protocols or scores interchangeable.

Baseline coverage

Reference methods help show what a model adds beyond simple controls. We track a null control and a conventional method for each protocol.

0 of 66 active baseline roles have published Rewire measurements in this release. Measurements on a selected protocol do not establish coverage of an entire suite.

Baseline status by linked protocol

Protocol coverage CSV · Model evaluation matrix · Source table · Release and checksums

Coverage is derived from release 2026-09-29-06401fd5b220. Source citations describe the original records; they do not validate an unreviewed baseline proposal. No results have been generated by this audit.

Run this benchmark

Choose a concrete protocol before running an evaluation. Its inputs, split and scoring rules determine which results can be compared.

Run a GlycanML task

Create the environment and run one of the benchmark's single-task or multi-task configurations.

Generate predictions and evaluate them. This recipe does not establish reproduction of a particular published score.

Dataset access
Prepared by the repository's configs.
Model and weights
None required; models are trained from the data.
Licences
Project licence: Apache-2.0. Upstream data licences are separate and unreported here.
Software
Python with the repository's conda environment and torchdrug.
Hardware
Not stated in the cited section. Several of these steps expect a GPU.
Required inputs and expected outputs

Inputs

  • A model configuration from the repository's config directory.

Outputs

  • Task scores written by the run script.

Execution steps

  1. 1. Install (Command line)

    Source reviewed; these instructions have not been executed by rewire.

    conda create -n torchdrug python=3.9
conda activate torchdrug

conda install --yes pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch
conda install --yes pyg pytorch-scatter pytorch-cluster -c pyg
pip install torchdrug
pip install pyyaml easydict scipy fair-esm
pip install dill biopandas biopython e3nn wandb tensorboard tensorboardX
pip install glycowork[draw]
    GlycanML: repository README · README.md at 9f392aa6, Installation, lines 30-38

  2. 2. Run a single task (Command line)

    Source reviewed; these instructions have not been executed by rewire.

    python scripts/run_single.py --config ./configs/single_task/$model/$yaml_config \
    --gpus [0] --seed 0
    GlycanML: repository README · README.md at 9f392aa6, Single-Task Learning, lines 80-81

  3. 3. Run the multi-task setting (Command line)

    Source reviewed; these instructions have not been executed by rewire.

    python scripts/run_single.py --config ./configs/multi_task/$model/$yaml_config \
    --gpus [0] --seed 0
    GlycanML: repository README · README.md at 9f392aa6, Multi-Task Learning, lines 98-99

Use your own model

Run your model locally and return predictions keyed by the input IDs. The evaluator supplies biological inputs without test labels and owns scoring. This interface is not a sandbox for model code.

Pass your existing prediction function into this adapter. Its output direction must match the selected protocol.

class MyModelAdapter:
    def __init__(self, score):
        self.score = score

    def predict(self, inputs):
        return {row["id"]: float(self.score(row)) for row in inputs}

# adapter = MyModelAdapter(your_prediction_function)
# report = rewirebench.run(prepared, adapter, output="runs/my-model")

Alternatively, generate a keyed prediction file in your existing model environment and use the score-only recipe. Your model code and weights do not need to be shared.

GlycanML: repository README · README.md at 9f392aa6
Scope and limitations
  • Quoted from the project's README and not executed by rewire, so the commands are evidence of what the project documents rather than a verified run.
  • The project may have changed since the pinned commit.
  • Task scores use different metrics and are not a single leaderboard.

Contribute a result for review. The library can submit an exported evaluation for private review when intake is open. Check the contribution page for access and sign-in.

Original repository instructions

Run this benchmark

Official installation and single-/multi-GPU training templates are available. A concrete task YAML and model must replace the placeholders; the prose refers to ./config while command templates use ./configs, so check the actual pinned path before constructing an executable recipe.

A maintained rewire runner has not been verified for this benchmark. Check data access, weights, licences, dependencies and hardware in the linked official documentation; requirements have not been fully extracted.

GlycanML/GlycanML / README.md · README.md lines 24–46 and 74–99 (Installation and Model Training)
Strengths, limitations and unresolved questions

Strengths and limitations

Strengths supported by sources

Limitations and conditions

  • Taxonomy, immunogenicity and glycosylation tasks hold out glycan motif clusters; interaction prediction holds out protein clusters. These boundaries do not imply that both proteins and glycans are unseen in the interaction task.
    Sourcesglycanml primary benchmark evidence · Sections 3.1–3.4 and 5.1; Tables 1 and 3
Profile review details

Primary paper and/or task implementation reviewed for the explicitly cited methodology claims. Scope-limited absence is recorded only after the documented source search; no model runs or independent reproduction.

Stable record: discovery-benchmark-glycanml

Specifications

Inputs, training, access and other details

Explanatory profile: source reviewed · Automated source review, 2026-09-16. Review applies to the cited claims; unresolved fields are listed below. Numerical results retain their own review status.

Data, procedure and scoring
PropertyDescription and evidence
DatasetsGlycan taxonomy, immunogenicity, glycosylation-type and protein–glycan interaction datasets.
Sources (5)GlycanML/GlycanML official source; GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml · Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction
SplitsTaxonomy, immunogenicity and glycosylation tasks use motif-cluster partitions. Protein–glycan interaction uses MMseqs2 protein clusters with threshold 0.5. Both allocate clusters 8:1:1; the two notions of held-out data are different.
Sourcesglycanml primary benchmark evidence · Sections 3.1–3.4
MetricsChecked single-task configurations use accuracy/MCC for taxonomy and glycosylation type, AUROC/AUPRC for immunogenicity, and MAE/RMSE/Spearman for protein–glycan interaction regression.
Sources (5)GlycanML/GlycanML official source; GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml · Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction
BaselinesSequence-model CNN, ResNet, LSTM and BERT configurations; graph-model GCN, RGCN, GAT, GIN, CompGCN and MPNN configurations.
Sources (5)GlycanML/GlycanML official source; GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml · Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction
Leakage controlsMotif-based cluster separation tests transfer to structurally different glycans. This is a glycan-structure control, not a claim that all organisms or source studies are held out.
Sourcesglycanml primary benchmark evidence · Sections 3.1–3.3; Table 1
UncertaintyEvery experiment uses seeds 0, 1 and 2; reported summaries are the mean and standard deviation over those three runs.
Sourcesglycanml primary benchmark evidence · Sections 3.1–3.4 and 5.1; Tables 1 and 3
Entity typeGlycan representation benchmark suite.
Sources (5)GlycanML/GlycanML official source; GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml · Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction
OrganismsTaxonomy tasks explicitly predict organism categories; species scope depends on the constituent dataset.
Sources (5)GlycanML/GlycanML official source; GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml · Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction
AssaysTaxonomy, immunogenicity, glycosylation-type and protein–glycan interaction annotations.
Sources (5)GlycanML/GlycanML official source; GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml · Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction
Allowed inputsGlycan sequence or graph representations and, for interaction tasks, paired protein data.
Sources (5)GlycanML/GlycanML official source; GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml · Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction
AdaptationSeparate single-task and multi-task training configurations are supplied.
Sources (5)GlycanML/GlycanML official source; GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml; GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml · Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction
Applicable tests and references

Applicability is distinct from a completed evaluation.

Evidence

Source checking verifies the cited claim or transcription. It does not establish independent reproduction.

Papers and result coverage

Last literature check: 2026-09-17. Primary-paper within-study comparison; numerical results not independently reproduced.

Paper or primary resourceVersionReference
GlycanML: A Multi-Task and Multi-Structure Benchmark for Glycan Machine LearningarXiv:2405.16206v1, 2024-05-25Read source
Historical gaps recorded on 2026-09-17

The catalogue now holds 323 result rows for this benchmark. A note below about pending extraction describes the state on 2026-09-17 and may since have been answered by a later batch. The result rows and their sources are the current record.

  • Paper table does not enumerate trained checkpoint hashes; source configuration names retained.
Search and extraction details

complete comparison extracted

Searches

  • GlycanML benchmark 2405.16206

Evidence locations

  • Sections 3–5
  • Tables 1–4

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.

81 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-09-29-06401fd5b220
Property and statementOriginal source and locationReview and provenance
Diagram caption
Conceptual procedure. Task variants and protocol versions retain their separate scoring conditions.
Individual claims		GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml

Original source ↗

Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; Sections 3.1–3.4

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: 9f392aa6f9c6d74a296a250199beb347923d04e0
Retrieved: 2026-09-16T20:42:46.930612+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary paper and/or task implementation reviewed for the explicitly cited methodology claims. Scope-limited absence is recorded only after the documented source search; no model runs or independent reproduction.

Field: attributes.profile.diagram.caption

Source artifact SHA-256: 5c92f79c530629642450ca4536cd864cccc3a626f8ecdfeaaf6fcd2ece356901

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

Inspected artifact

Diagram caption
Conceptual procedure. Task variants and protocol versions retain their separate scoring conditions.
Individual claims		GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml

Original source ↗

Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; Sections 3.1–3.4

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: 9f392aa6f9c6d74a296a250199beb347923d04e0
Retrieved: 2026-09-16T20:42:47.417430+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary paper and/or task implementation reviewed for the explicitly cited methodology claims. Scope-limited absence is recorded only after the documented source search; no model runs or independent reproduction.

Field: attributes.profile.diagram.caption

Source artifact SHA-256: 7cb8f9e774ab0f9d66536b2db47bb236214275ee08d4de95540e4e51d0129639

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

Inspected artifact

Diagram caption
Conceptual procedure. Task variants and protocol versions retain their separate scoring conditions.
Individual claims		GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml

Original source ↗

Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; Sections 3.1–3.4

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: 9f392aa6f9c6d74a296a250199beb347923d04e0
Retrieved: 2026-09-16T20:42:47.161322+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary paper and/or task implementation reviewed for the explicitly cited methodology claims. Scope-limited absence is recorded only after the documented source search; no model runs or independent reproduction.

Field: attributes.profile.diagram.caption

Source artifact SHA-256: 82f43f0d45085132ecd6e6dff1b95bf7eb20ad5650713fc8ad0a42d0bbba5a94

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

Inspected artifact

Diagram caption
Conceptual procedure. Task variants and protocol versions retain their separate scoring conditions.
Individual claims		GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml

Original source ↗

Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; Sections 3.1–3.4

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: 9f392aa6f9c6d74a296a250199beb347923d04e0
Retrieved: 2026-09-16T20:42:46.675170+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary paper and/or task implementation reviewed for the explicitly cited methodology claims. Scope-limited absence is recorded only after the documented source search; no model runs or independent reproduction.

Field: attributes.profile.diagram.caption

Source artifact SHA-256: 474995bb150c76479b0c99b6f0b2de49f5b6ad626eae1751e2f53774a4fbf38e

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

Inspected artifact

Diagram caption
Conceptual procedure. Task variants and protocol versions retain their separate scoring conditions.
Individual claims		glycanml primary benchmark evidence

Original source ↗

Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; Sections 3.1–3.4

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: 2405.16206v1
Retrieved: 2026-09-16T21:04:56.017006+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary paper and/or task implementation reviewed for the explicitly cited methodology claims. Scope-limited absence is recorded only after the documented source search; no model runs or independent reproduction.

Field: attributes.profile.diagram.caption

Source artifact SHA-256: 9ba3db678b4550898a612b42e8832bf9dee40935090fc4d969ba8f6ac5106979

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

Inspected artifact

Diagram caption
Conceptual procedure. Task variants and protocol versions retain their separate scoring conditions.
Individual claims		GlycanML/GlycanML official source

Original source ↗

Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; Sections 3.1–3.4

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: 9f392aa6f9c6d74a296a250199beb347923d04e0
Retrieved: 2026-09-16T10:30:24.094166+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary paper and/or task implementation reviewed for the explicitly cited methodology claims. Scope-limited absence is recorded only after the documented source search; no model runs or independent reproduction.

Field: attributes.profile.diagram.caption

Source artifact SHA-256: 5237cd1af3f1d7b9cf07cfe3cf722987bb27e10ee356ec108584943446bbb836

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

Diagram steps
  • Allowed inputs: Glycan sequence or graph representations and, for interaction tasks, paired protein data.
  • Splits: Taxonomy, immunogenicity and glycosylation tasks use motif-cluster partitions. Protein–glycan interaction uses MMseqs2 protein clusters with threshold 0.5. Both allocate clusters 8:1:1; the two notions of held-out data are different.
  • Metrics: Checked single-task configurations use accuracy/MCC for taxonomy and glycosylation type, AUROC/AUPRC for immunogenicity, and MAE/RMSE/Spearman for protein–glycan interaction regression.
Individual claims		GlycanML/GlycanML configs/single_task/BERT/immunogenicity_BERT.yaml

Original source ↗

Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; Sections 3.1–3.4

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: 9f392aa6f9c6d74a296a250199beb347923d04e0
Retrieved: 2026-09-16T20:42:46.930612+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary paper and/or task implementation reviewed for the explicitly cited methodology claims. Scope-limited absence is recorded only after the documented source search; no model runs or independent reproduction.

Field: attributes.profile.diagram.steps

Source artifact SHA-256: 5c92f79c530629642450ca4536cd864cccc3a626f8ecdfeaaf6fcd2ece356901

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

Inspected artifact

Diagram steps
  • Allowed inputs: Glycan sequence or graph representations and, for interaction tasks, paired protein data.
  • Splits: Taxonomy, immunogenicity and glycosylation tasks use motif-cluster partitions. Protein–glycan interaction uses MMseqs2 protein clusters with threshold 0.5. Both allocate clusters 8:1:1; the two notions of held-out data are different.
  • Metrics: Checked single-task configurations use accuracy/MCC for taxonomy and glycosylation type, AUROC/AUPRC for immunogenicity, and MAE/RMSE/Spearman for protein–glycan interaction regression.
Individual claims		GlycanML/GlycanML configs/single_task/BERT/interaction_BERT.yaml

Original source ↗

Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; Sections 3.1–3.4

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: 9f392aa6f9c6d74a296a250199beb347923d04e0
Retrieved: 2026-09-16T20:42:47.417430+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary paper and/or task implementation reviewed for the explicitly cited methodology claims. Scope-limited absence is recorded only after the documented source search; no model runs or independent reproduction.

Field: attributes.profile.diagram.steps

Source artifact SHA-256: 7cb8f9e774ab0f9d66536b2db47bb236214275ee08d4de95540e4e51d0129639

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

Inspected artifact

Diagram steps
  • Allowed inputs: Glycan sequence or graph representations and, for interaction tasks, paired protein data.
  • Splits: Taxonomy, immunogenicity and glycosylation tasks use motif-cluster partitions. Protein–glycan interaction uses MMseqs2 protein clusters with threshold 0.5. Both allocate clusters 8:1:1; the two notions of held-out data are different.
  • Metrics: Checked single-task configurations use accuracy/MCC for taxonomy and glycosylation type, AUROC/AUPRC for immunogenicity, and MAE/RMSE/Spearman for protein–glycan interaction regression.
Individual claims		GlycanML/GlycanML configs/single_task/BERT/link_BERT.yaml

Original source ↗

Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; Sections 3.1–3.4

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: 9f392aa6f9c6d74a296a250199beb347923d04e0
Retrieved: 2026-09-16T20:42:47.161322+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary paper and/or task implementation reviewed for the explicitly cited methodology claims. Scope-limited absence is recorded only after the documented source search; no model runs or independent reproduction.

Field: attributes.profile.diagram.steps

Source artifact SHA-256: 82f43f0d45085132ecd6e6dff1b95bf7eb20ad5650713fc8ad0a42d0bbba5a94

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

Inspected artifact

Diagram steps
  • Allowed inputs: Glycan sequence or graph representations and, for interaction tasks, paired protein data.
  • Splits: Taxonomy, immunogenicity and glycosylation tasks use motif-cluster partitions. Protein–glycan interaction uses MMseqs2 protein clusters with threshold 0.5. Both allocate clusters 8:1:1; the two notions of held-out data are different.
  • Metrics: Checked single-task configurations use accuracy/MCC for taxonomy and glycosylation type, AUROC/AUPRC for immunogenicity, and MAE/RMSE/Spearman for protein–glycan interaction regression.
Individual claims		GlycanML/GlycanML configs/single_task/BERT/species_BERT.yaml

Original source ↗

Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; Sections 3.1–3.4

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: 9f392aa6f9c6d74a296a250199beb347923d04e0
Retrieved: 2026-09-16T20:42:46.675170+00:00

source checked

automated source review · 2026-09-16

Audit details

Primary paper and/or task implementation reviewed for the explicitly cited methodology claims. Scope-limited absence is recorded only after the documented source search; no model runs or independent reproduction.

Field: attributes.profile.diagram.steps

Source artifact SHA-256: 474995bb150c76479b0c99b6f0b2de49f5b6ad626eae1751e2f53774a4fbf38e

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

Inspected artifact

All evidence (CSV)All evidence (JSONL)How to read the evidence

Sources and history

View linked audit checks and correction history

Release 2026-09-29-06401fd5b220 · Record review: discovered

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

Stable ID: discovery-benchmark-glycanml

areas
glycomics
entity level
suite
scope note
Specialist molecular or omics evaluation; protocol details require review before numerical comparison.
task
Glycan properties, taxonomy and molecular interactions
version
Not reported
comparison panels
id: glycanml-v1-single-domain; title: GlycanML Domain · Table 3, p. 8; protocol id: paper-protocol-b51570e2bf69ba6542; dataset id: paper-dataset-ce710f303c24cb91c2; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-075ca33548bf92ef7e; paper-result-0b5a6c78c01691f3ee; paper-result-d0bee0d884f958e49b; paper-result-4543142c4187cbea18; paper-result-ea2bf935a59e8b5693; paper-result-a4c79e5939d76f3983; paper-result-c3d21b293a8aa00c67; paper-result-5c30f19e8d1c4b91ca; paper-result-2aed0631283783ea66; paper-result-22b89e6bc2ccf2fe1b; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 3, p. 8, Domain; context: Author-reported single-task models on the same held-out task; not a cross-paper ranking. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.; Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.; Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-single-kingdom; title: GlycanML Kingdom · Table 3, p. 8; protocol id: paper-protocol-468fdd0f4237c041d5; dataset id: paper-dataset-d7ac4e0285fbaefcab; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-09a8ee00d327d0e580; paper-result-f088c48c3867898d15; paper-result-362d36486fd109e109; paper-result-a50f7f1f26ebaa5d42; paper-result-45611bc39c803891d0; paper-result-f99f24cc69642bce07; paper-result-a0c917c8b7a0bf0388; paper-result-410bae2881489d32dd; paper-result-8dea7f67a6307f6e1f; paper-result-98203c7be8a19baf6c; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 3, p. 8, Kingdom; context: Author-reported single-task models on the same held-out task; not a cross-paper ranking. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.; Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.; Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-single-phylum; title: GlycanML Phylum · Table 3, p. 8; protocol id: paper-protocol-a67c6f1eaf73494953; dataset id: paper-dataset-13c71587fb39174bb1; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-86a3427fdae5f397c2; paper-result-c861ec21227123bf0b; paper-result-c66956e19c73fa9abf; paper-result-c1f20eec6e0881c783; paper-result-82156f189d0b3f8cd9; paper-result-3e5677e0c95596dcd7; paper-result-34a6e2a047a75ce39d; paper-result-2dfea46ca6f79b2ec1; paper-result-778ad7ec5934e369ed; paper-result-c99e02301c02669cc1; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 3, p. 8, Phylum; context: Author-reported single-task models on the same held-out task; not a cross-paper ranking. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.; Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.; Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-single-class; title: GlycanML Class · Table 3, p. 8; protocol id: paper-protocol-a8799c4d8546b3ff18; dataset id: paper-dataset-f9b7538618a3e1ef5b; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-32ae4afd9a4c6cc89b; paper-result-da7afa1436a5e1fae4; paper-result-504992911a85e182cc; paper-result-c459dd0862f1e48ee5; paper-result-2b302ea797608740bf; paper-result-966ff1085af74c7b89; paper-result-524b419436ec9d8192; paper-result-d9c96e1c04f27b0d48; paper-result-ed1c70949ef07f3093; paper-result-233bc2d9c1d6575a1e; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 3, p. 8, Class; context: Author-reported single-task models on the same held-out task; not a cross-paper ranking. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.; Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.; Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-single-order; title: GlycanML Order · Table 3, p. 8; protocol id: paper-protocol-fb9190db65a5f72e10; dataset id: paper-dataset-4298ac38e85e6f8048; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-abee54829f2ea08080; paper-result-7281a16570ce97b5cb; paper-result-cefa58a1f691f86ec4; paper-result-35cbadb92d47ea3c60; paper-result-3f3f4af93e3f57eacd; paper-result-0d4e92bac0d910c607; paper-result-3a9fc20851e90815ea; paper-result-d8d2084714e258967b; paper-result-ac7e4ed00d0e9515da; paper-result-af527f9d744c668717; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 3, p. 8, Order; context: Author-reported single-task models on the same held-out task; not a cross-paper ranking. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.; Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.; Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-single-family; title: GlycanML Family · Table 3, p. 8; protocol id: paper-protocol-8dca9eabea26dee766; dataset id: paper-dataset-8d5d79e4a6a070bfff; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-09e347c7674f561706; paper-result-8bbf97e6e23787da99; paper-result-f5c63db202d546dd8a; paper-result-941024aaffa398c454; paper-result-32c123a2e27ebbe79e; paper-result-4a140ae1cf5519925c; paper-result-76cbc3d2466076b944; paper-result-42fcbda81a387d226d; paper-result-e4ad7f8d96df00ea4b; paper-result-1337bae859236e6dcf; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 3, p. 8, Family; context: Author-reported single-task models on the same held-out task; not a cross-paper ranking. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.; Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.; Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-single-genus; title: GlycanML Genus · Table 3, p. 8; protocol id: paper-protocol-d44b40251c33ecbea1; dataset id: paper-dataset-df6da272a0a4d8de8f; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-ca073eb7b8ec09d81b; paper-result-ec1560d4f26c854db8; paper-result-012e11422e843b97f7; paper-result-6d2f9510b6bb3023ca; paper-result-6a9cd20bbb10eaeb7e; paper-result-5e1004788b7b7c0281; paper-result-f3d9c9bc60ec3b9cc7; paper-result-e30b9280e6a5a6fd1c; paper-result-fe8185d4fa2fa84e4e; paper-result-fae731c13691351a52; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 3, p. 8, Genus; context: Author-reported single-task models on the same held-out task; not a cross-paper ranking. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.; Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.; Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-single-species; title: GlycanML Species · Table 3, p. 8; protocol id: paper-protocol-0d66c7cabff9d799d5; dataset id: paper-dataset-4c48251022ccb53313; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-dc5501dc1954b54a36; paper-result-825d6b6597e463e7cc; paper-result-fedc9b0d7fadb8f74e; paper-result-19f2505d003bc76dcc; paper-result-e9a6b117f8a49863be; paper-result-33876cb9abd2083b08; paper-result-fad4c5f6d2c179bb3f; paper-result-77f9c63846886893da; paper-result-723b93ccacba688198; paper-result-e74c30428cded73273; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 3, p. 8, Species; context: Author-reported single-task models on the same held-out task; not a cross-paper ranking. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.; Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.; Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-single-immuno; title: GlycanML Immuno · Table 3, p. 8; protocol id: paper-protocol-705c3995af4e74e425; dataset id: paper-dataset-dfa9efa0038fe35cf1; metric: AUPRC; unit: dimensionless; direction: higher; result ids: paper-result-daed94f39b26d6bf9f; paper-result-14cf9ecb36e308a19c; paper-result-0f0c6ac7d2490faded; paper-result-a6f2c25189be55752b; paper-result-c3c413b66dd284a72c; paper-result-cfb7d3096516691659; paper-result-c17bcd75195fbd91e3; paper-result-76c0f9b5fb05a8b66f; paper-result-316c3afd5495e9cc25; paper-result-8afa28bb629f98ba97; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 3, p. 8, Immuno; context: Author-reported single-task models on the same held-out task; not a cross-paper ranking. Taxonomy, immunogenicity and glycosylation: motif-frequency K-means cluster allocation 8:1:1. Interaction: MMseqs2 protein clusters at minimum identity 0.5 allocated 8:1:1. Exact counts are attached to each column.; caveats: Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.; Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.; Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-single-glycos; title: GlycanML Glycos · Table 3, p. 8; protocol id: paper-protocol-1e0e586e8edab0dc26; dataset id: paper-dataset-e797e17c76c1278706; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-ee2fc7d3f24c74d26a; paper-result-cd7d5e177d170e4ce3; paper-result-c7fcf3af8b768565b7; paper-result-e2f3edb52244aed4ea; paper-result-edb19c124f6d9e2398; paper-result-ebfebb353d4b0a9e5a; paper-result-04b741864eefb0125e; paper-result-00ca74108e1dd8a295; paper-result-7ef41afc692b853ffd; paper-result-c4ba6aef76ee19b8d5; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 3, p. 8, Glycos; context: Author-reported single-task models on the same held-out task; not a cross-paper ranking. Taxonomy, immunogenicity and glycosylation: motif-frequency K-means cluster allocation 8:1:1. Interaction: MMseqs2 protein clusters at minimum identity 0.5 allocated 8:1:1. Exact counts are attached to each column.; caveats: Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.; Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.; Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-single-interaction; title: GlycanML Interaction · Table 3, p. 8; protocol id: paper-protocol-e914e4149f3b3c8185; dataset id: paper-dataset-cc0cb44a535d3e0e7a; metric: Spearman rho; unit: dimensionless; direction: higher; result ids: paper-result-16108af1186bb07643; paper-result-b40c9e0218e574eca5; paper-result-9284b039f056e2a302; paper-result-cc74f69913e5a2aaa7; paper-result-cd20816595a9f309d0; paper-result-37e29e684dddd82a16; paper-result-6bc53e8c78556b6241; paper-result-e3e3952c325b6ef0a2; paper-result-f5c722997fa24f0560; paper-result-9865abcc65a28a6bed; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 3, p. 8, Interaction; context: Author-reported single-task models on the same held-out task; not a cross-paper ranking. Taxonomy, immunogenicity and glycosylation: motif-frequency K-means cluster allocation 8:1:1. Interaction: MMseqs2 protein clusters at minimum identity 0.5 allocated 8:1:1. Exact counts are attached to each column.; caveats: Taxonomic levels are distinct tasks; do not combine raw accuracy across levels.; Protein–glycan scores evaluate combined glycan encoders plus ESM-1b/MLP, not an ESM-1b standalone checkpoint.; Cluster manifests and trained checkpoint hashes are not enumerated in the paper table.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-0-domain; title: GlycanML Domain · Table 4, p. 9; protocol id: paper-protocol-b51570e2bf69ba6542; dataset id: paper-dataset-ce710f303c24cb91c2; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-075ca33548bf92ef7e; paper-result-fdb107a7ebf70f41b9; paper-result-0f87601e912c4389c4; paper-result-cea1fd4ede7da082f0; paper-result-a1d4fa5ae3eeb76bef; paper-result-425410a32e12403336; paper-result-7d6bfd8ae273536a43; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Domain; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-0-kingdom; title: GlycanML Kingdom · Table 4, p. 9; protocol id: paper-protocol-468fdd0f4237c041d5; dataset id: paper-dataset-d7ac4e0285fbaefcab; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-09a8ee00d327d0e580; paper-result-b8f593d66e81d8176a; paper-result-8ac1e91e821535aaff; paper-result-1c68188144c78ba223; paper-result-59aaa70a642c016c71; paper-result-3a2676e4651199a4c7; paper-result-5f6263a893e28a7406; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Kingdom; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-0-phylum; title: GlycanML Phylum · Table 4, p. 9; protocol id: paper-protocol-a67c6f1eaf73494953; dataset id: paper-dataset-13c71587fb39174bb1; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-86a3427fdae5f397c2; paper-result-8c1657f118a2dd808c; paper-result-ad0d1dd5c3cfc326af; paper-result-9dc690711bc090f19a; paper-result-a0214b935eba01b7a7; paper-result-43fa9c20d6d6b41325; paper-result-aec7ac1f09709fe2e2; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Phylum; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-0-class; title: GlycanML Class · Table 4, p. 9; protocol id: paper-protocol-a8799c4d8546b3ff18; dataset id: paper-dataset-f9b7538618a3e1ef5b; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-32ae4afd9a4c6cc89b; paper-result-0bf3b2d166c172f494; paper-result-5cf336e78e8b9646f7; paper-result-3aa13c82aac4f30398; paper-result-750f1db97156302340; paper-result-50d6f0ba85fc1e572d; paper-result-5e2c55226c432a927b; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Class; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-0-order; title: GlycanML Order · Table 4, p. 9; protocol id: paper-protocol-fb9190db65a5f72e10; dataset id: paper-dataset-4298ac38e85e6f8048; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-abee54829f2ea08080; paper-result-a299bb2064e93cd095; paper-result-56157f637f3dc28dde; paper-result-d72371bae7104d40bb; paper-result-6692b5f2ab128e869f; paper-result-b20abedfaf179beb0f; paper-result-b1a73f13e672f9f010; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Order; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-0-family; title: GlycanML Family · Table 4, p. 9; protocol id: paper-protocol-8dca9eabea26dee766; dataset id: paper-dataset-8d5d79e4a6a070bfff; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-09e347c7674f561706; paper-result-89df39b51fc8f73a18; paper-result-3710d0e01f022a8f58; paper-result-76b9f005da338ff5f4; paper-result-97cf3af868ea202c20; paper-result-0f26b2626d93d81e1a; paper-result-45cd9dbf505c6c3edf; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Family; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-0-genus; title: GlycanML Genus · Table 4, p. 9; protocol id: paper-protocol-d44b40251c33ecbea1; dataset id: paper-dataset-df6da272a0a4d8de8f; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-ca073eb7b8ec09d81b; paper-result-658a0bde785ae82791; paper-result-ba1188eafda83f2dff; paper-result-8ef3992de76e028698; paper-result-72ccde0fe45e939979; paper-result-4dc4b5b8e2b2f51f0d; paper-result-e2393930743e0fe2a3; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Genus; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-0-species; title: GlycanML Species · Table 4, p. 9; protocol id: paper-protocol-0d66c7cabff9d799d5; dataset id: paper-dataset-4c48251022ccb53313; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-dc5501dc1954b54a36; paper-result-3decc92d3c4b47daaa; paper-result-efe8061d8b717b3d06; paper-result-994aa15ef63ab963bd; paper-result-af7e44b66703958045; paper-result-7741da71a6bffbfe05; paper-result-2a425531358cee050f; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Species; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-0-mean-acc; title: GlycanML Mean Acc (%) · Table 4, p. 9; protocol id: paper-protocol-f6058db65ace02f920; dataset id: paper-dataset-3b08dc110e25f5011e; metric: mean accuracy across eight taxonomy tasks; unit: percent; direction: higher; result ids: paper-result-3b8254d638b7cf840c; paper-result-26353885331e3500b0; paper-result-14f9374f43c9937375; paper-result-37820120325451b083; paper-result-010096d9a694ae3520; paper-result-96a241380ae6ba882a; paper-result-222b0816ea58c393cd; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Mean Acc (%); context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-1-domain; title: GlycanML Domain · Table 4, p. 9; protocol id: paper-protocol-b51570e2bf69ba6542; dataset id: paper-dataset-ce710f303c24cb91c2; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-2aed0631283783ea66; paper-result-944ec4d9e5d591defb; paper-result-976c07b788d0c976b5; paper-result-db9fa9da2f1bc0b208; paper-result-f0d520d129d1982892; paper-result-abafa8cbc9111307ef; paper-result-c1ebf0846e005b0ea7; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Domain; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-1-kingdom; title: GlycanML Kingdom · Table 4, p. 9; protocol id: paper-protocol-468fdd0f4237c041d5; dataset id: paper-dataset-d7ac4e0285fbaefcab; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-8dea7f67a6307f6e1f; paper-result-89b973f6b24d20d70b; paper-result-7c4ad10784c12bc139; paper-result-853910c368572b5ab8; paper-result-0fd9b263d4d481168c; paper-result-a189d3bce6f0744a64; paper-result-c1581b8b54883511ea; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Kingdom; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-1-phylum; title: GlycanML Phylum · Table 4, p. 9; protocol id: paper-protocol-a67c6f1eaf73494953; dataset id: paper-dataset-13c71587fb39174bb1; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-778ad7ec5934e369ed; paper-result-d60a1868d6cac6686d; paper-result-3faf014c06fdb5b338; paper-result-6a836283cecd69f216; paper-result-077aee2f6ad33075b7; paper-result-26eb6ff8bcba531860; paper-result-8776fd8149834a3faf; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Phylum; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-1-class; title: GlycanML Class · Table 4, p. 9; protocol id: paper-protocol-a8799c4d8546b3ff18; dataset id: paper-dataset-f9b7538618a3e1ef5b; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-ed1c70949ef07f3093; paper-result-c4239ca0a78748ef44; paper-result-d6364ea9b2a3f8fdaf; paper-result-34b64c1f78f1731268; paper-result-8f14f53fefa6b454b9; paper-result-42546bacf62f9b3a69; paper-result-7a11fb3230b1f4dcec; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Class; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-1-order; title: GlycanML Order · Table 4, p. 9; protocol id: paper-protocol-fb9190db65a5f72e10; dataset id: paper-dataset-4298ac38e85e6f8048; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-ac7e4ed00d0e9515da; paper-result-e6c210784feca5b6d4; paper-result-f5553ce1aa89330367; paper-result-57e2eb21a87ce0ad82; paper-result-f8ab2a4236081aaa2d; paper-result-200cf342b98083f8eb; paper-result-6abd4dddb7fdcb0691; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Order; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-1-family; title: GlycanML Family · Table 4, p. 9; protocol id: paper-protocol-8dca9eabea26dee766; dataset id: paper-dataset-8d5d79e4a6a070bfff; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-e4ad7f8d96df00ea4b; paper-result-7e5eb5e5def1997d2a; paper-result-6189ade0e2553e6a5f; paper-result-850116eec75fde4e16; paper-result-a90400ff944a9dba82; paper-result-4089bd143f1ec302b3; paper-result-b691a9c8c205144c7f; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Family; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-1-genus; title: GlycanML Genus · Table 4, p. 9; protocol id: paper-protocol-d44b40251c33ecbea1; dataset id: paper-dataset-df6da272a0a4d8de8f; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-fe8185d4fa2fa84e4e; paper-result-14bd301ffa5bdd2ba0; paper-result-6ae0a365e6b33ffa9b; paper-result-fc6174c38a0298468e; paper-result-410dcdb82f6d6c3b0f; paper-result-54ce873648167ddf7f; paper-result-aa50763240e086a7b0; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Genus; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-1-species; title: GlycanML Species · Table 4, p. 9; protocol id: paper-protocol-0d66c7cabff9d799d5; dataset id: paper-dataset-4c48251022ccb53313; metric: accuracy; unit: percent; direction: higher; result ids: paper-result-723b93ccacba688198; paper-result-ec8c144273ccc0a850; paper-result-928828d65bf0d96282; paper-result-5f765144a6d6d5bdb4; paper-result-dc366357aed61c1324; paper-result-618be13e9cdea80f33; paper-result-e3de6f13924b3f0956; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Species; context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17; id: glycanml-v1-mtl-1-mean-acc; title: GlycanML Mean Acc (%) · Table 4, p. 9; protocol id: paper-protocol-f6058db65ace02f920; dataset id: paper-dataset-3b08dc110e25f5011e; metric: mean accuracy across eight taxonomy tasks; unit: percent; direction: higher; result ids: paper-result-5ebd480b2815e83e3d; paper-result-13c2836aee0f4143ea; paper-result-19abc6d9c3e6eaf449; paper-result-16f47917baa3903d08; paper-result-2d85ed1325988db66d; paper-result-31eba8c1c674dcaf66; paper-result-ecd8e5a52b2ee62bca; source ids: expansion-p3-glycanml-2405-16206v1; source locator: Table 4, p. 9, Mean Acc (%); context: Fixed backbone; compare adaptation strategies, not standalone foundation models. SugarBase taxonomy: motif-frequency K-means cluster allocation 8:1:1.; caveats: Mean accuracy is an unweighted aggregate of eight taxonomic tasks as reported; keep separate from every individual level.; The six MTL approaches change training, so do not label their scores as the base encoder alone.; review: method: automated_source_review; date: 2026-09-17
benchmark research
review date: 2026-09-17; status: complete_comparison_extracted; primary sources: expansion-p3-glycanml-2405-16206v1; inspected locators: Sections 3–5; Tables 1–4; searched queries: GlycanML benchmark 2405.16206; gaps: Paper table does not enumerate trained checkpoint hashes; source configuration names retained.; claim scope: Primary-paper within-study comparison; numerical results not independently reproduced.
historical missing metadata
dataset release: unextracted; metric implementation: unextracted; split manifest: unextracted; version: unextracted
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.
entity classification
review date: 2026-09-17; rationale: The cited profile describes a collection of evaluation tasks or protocols; retain it as the top-level benchmark suite. Its datasets and individual protocols remain separate records.; source ids: src-discovery-glycanml-glycanml; evidence-benchmark-glycanml-bert-species-bert-yaml; evidence-benchmark-glycanml-bert-immunogenicity-bert-yaml; evidence-benchmark-glycanml-bert-link-bert-yaml; evidence-benchmark-glycanml-bert-interaction-bert-yaml; source locator: Pinned README: Introduction; Experiment configurations; leaderboard; pinned single-task BERT configurations for species, link, immunogenicity and interaction; ambiguities: None recorded
run documentation
record id: discovery-benchmark-glycanml; source ids: run-doc-glycanml-readme-md-9f392aa6; status: official_documentation_linked; summary: Official installation and single-/multi-GPU training templates are available. A concrete task YAML and model must replace the placeholders; the prose refers to ./config while command templates use ./configs, so check the actual pinned path before constructing an executable recipe.; source locator: README.md lines 24–46 and 74–99 (Installation and Model Training)
run recipes
id: glycanml-official; protocol id: discovery-benchmark-glycanml; version: 9f392aa6f9c6d74a296a250199beb347923d04e0; title: Run a GlycanML task; purpose: generate_and_evaluate; summary: Create the environment and run one of the benchmark's single-task or multi-task configurations.; inputs: A model configuration from the repository's config directory.; outputs: Task scores written by the run script.; requirements: data: Prepared by the repository's configs.; weights: None required; models are trained from the data.; licence: Project licence: Apache-2.0. Upstream data licences are separate and unreported here.; software: Python with the repository's conda environment and torchdrug.; hardware: Not stated in the cited section. Several of these steps expect a GPU.; instructions: runtime: command_line; title: Install; code: conda create -n torchdrug python=3.9 conda activate torchdrug conda install --yes pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch conda install --yes pyg pytorch-scatter pytorch-cluster -c pyg pip install torchdrug pip install pyyaml easydict scipy fair-esm pip install dill biopandas biopython e3nn wandb tensorboard tensorboardX pip install glycowork[draw]; status: source_reviewed_not_executed; source ids: project-recipe-glycanml-9f392aa6; source locator: README.md at 9f392aa6, Installation, lines 30-38; runtime: command_line; title: Run a single task; code: python scripts/run_single.py --config ./configs/single_task/$model/$yaml_config \ --gpus [0] --seed 0; status: source_reviewed_not_executed; source ids: project-recipe-glycanml-9f392aa6; source locator: README.md at 9f392aa6, Single-Task Learning, lines 80-81; runtime: command_line; title: Run the multi-task setting; code: python scripts/run_single.py --config ./configs/multi_task/$model/$yaml_config \ --gpus [0] --seed 0; status: source_reviewed_not_executed; source ids: project-recipe-glycanml-9f392aa6; source locator: README.md at 9f392aa6, Multi-Task Learning, lines 98-99; limitations: Quoted from the project's README and not executed by rewire, so the commands are evidence of what the project documents rather than a verified run.; The project may have changed since the pinned commit.; Task scores use different metrics and are not a single leaderboard.; source ids: project-recipe-glycanml-9f392aa6; source locator: README.md at 9f392aa6
Related records

Suggest a correction