Protocol

OGBl-ppa PPI link prediction, Hits@100 (Table 5)

Evidence concern: excluded from comparisons

The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6

PPI link prediction on OGBl-ppa with protein embeddings as GNN node features; Hits@100 as printed in Table 5.

77 evaluations · 77 results

Overview

An explanatory profile has not yet been reviewed for this record.

Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.

77 recorded evaluations, 77 metric rows. A comparison chart has not yet been validated for these results. The table retains the individual findings and their sources.

View coverage and remaining gaps across all benchmarks

Results

Results are available, but no reviewed comparison panel is linked in this release.

All evaluations

77 evaluations · 77 results. Different protocols are not a single leaderboard.

Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) · Dataset: OGBl-ppa PPI graph (embedding study)

Sorted by Hits at 100 (OGBl-ppa PPI link prediction) (higher is better). The best value in each column is highlighted. Decimals are rounded for display; each value links to the printed value and its source.

Tested configuration	Hits at 100 (OGBl-ppa PPI link prediction)
NGNN with GCN combined with SEAL with ComplEx protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.627 (best in column)
NGNN with GraphSAGE combined with SEAL with RDF2Vec protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.599
NGNN with GCN combined with SEAL with RDF2Vec protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.594
NGNN with GraphSAGE combined with SEAL with ComplEx protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.572
NGNN with GraphSAGE combined with SEAL with randomly initialised node features (baseline) (PPI embeddings study)Evaluation	0.569
NGNN with GraphSAGE combined with SEAL with TransE protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.55
NGNN with GCN combined with SEAL with randomly initialised node features (baseline) (PPI embeddings study)Evaluation	0.512
NGNN with GCN combined with SEAL with majority vote across the representations (PPI embeddings study)Evaluation	0.51
NGNN with GCN combined with SEAL with ProGen2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.51
NGNN with GCN combined with SEAL with PiFold protein embeddings (structure) (PPI embeddings study)Evaluation	0.503
NGNN with GraphSAGE combined with SEAL with majority vote across the representations (PPI embeddings study)Evaluation	0.486
NGNN with GraphSAGE with PiFold protein embeddings (structure) (PPI embeddings study)Evaluation	0.485
NGNN with GCN combined with SEAL with TransE protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.478
NGNN with GraphSAGE combined with SEAL with PiFold protein embeddings (structure) (PPI embeddings study)Evaluation	0.469
NGNN with GCN combined with SEAL with ESM-IF1 protein embeddings (structure) (PPI embeddings study)Evaluation	0.469
NGNN with GCN combined with SEAL with ProtTrans protein embeddings (sequence) (PPI embeddings study)Evaluation	0.466
NGNN with GraphSAGE combined with SEAL with ProGen2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.463
NGNN with GCN with PiFold protein embeddings (structure) (PPI embeddings study)Evaluation	0.457
NGNN with GraphSAGE combined with SEAL with ESM-IF1 protein embeddings (structure) (PPI embeddings study)Evaluation	0.44
NGNN with GraphSAGE combined with SEAL with ProtTrans protein embeddings (sequence) (PPI embeddings study)Evaluation	0.438
NGNN with GraphSAGE combined with SEAL with ESM-2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.437
NGNN with GraphSAGE combined with SEAL with ProstT5 protein embeddings (structure) (PPI embeddings study)Evaluation	0.437
NGNN with GraphSAGE with RDF2Vec protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.434
NGNN with GCN with RDF2Vec protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.429
NGNN with GraphSAGE with ComplEx protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.428
NGNN with GCN with ComplEx protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.428
NGNN with GraphSAGE with ESM-IF1 protein embeddings (structure) (PPI embeddings study)Evaluation	0.422
NGNN with GCN combined with SEAL with ESM-2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.419
NGNN with GCN combined with SEAL with ProstT5 protein embeddings (structure) (PPI embeddings study)Evaluation	0.413
NGNN with GraphSAGE with TransE protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.41
NGNN with GCN with TransE protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.407
NGNN with GraphSAGE with ProGen2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.4
NGNN with GCN with majority vote across the representations (PPI embeddings study)Evaluation	0.358
NGNN with GCN with ProGen2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.353
NGNN with GCN with ESM-IF1 protein embeddings (structure) (PPI embeddings study)Evaluation	0.351
NGNN with GraphSAGE with majority vote across the representations (PPI embeddings study)Evaluation	0.331
NGNN with GCN with ESM-2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.311
NGNN with GCN with ProstT5 protein embeddings (structure) (PPI embeddings study)Evaluation	0.309
NGNN with GraphSAGE with ESM-2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.282
GraphSAGE with ComplEx protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.276
NGNN with GCN with ProtTrans protein embeddings (sequence) (PPI embeddings study)Evaluation	0.274
GCN with RDF2Vec protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.255
NGNN with GraphSAGE with randomly initialised node features (baseline) (PPI embeddings study)Evaluation	0.254
NGNN with GraphSAGE with ProtTrans protein embeddings (sequence) (PPI embeddings study)Evaluation	0.248
GCN with ComplEx protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.245
GraphSAGE with ESM-2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.242
GCN with ProGen2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.218
MLP on concatenated protein embeddings (protein-only baseline, no network) with RDF2Vec protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.21
GCN with majority vote across the representations (PPI embeddings study)Evaluation	0.21
GraphSAGE with PiFold protein embeddings (structure) (PPI embeddings study)Evaluation	0.209
GraphSAGE with RDF2Vec protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.198
MLP on concatenated protein embeddings (protein-only baseline, no network) with ComplEx protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.197
MLP on concatenated protein embeddings (protein-only baseline, no network) with TransE protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.192
GCN with PiFold protein embeddings (structure) (PPI embeddings study)Evaluation	0.189
GCN with ESM-IF1 protein embeddings (structure) (PPI embeddings study)Evaluation	0.185
GraphSAGE with ProGen2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.185
GCN with ProtTrans protein embeddings (sequence) (PPI embeddings study)Evaluation	0.168
GraphSAGE with majority vote across the representations (PPI embeddings study)Evaluation	0.167
GCN with ProstT5 protein embeddings (structure) (PPI embeddings study)Evaluation	0.162
GCN with ESM-2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.16
GraphSAGE with ESM-IF1 protein embeddings (structure) (PPI embeddings study)Evaluation	0.155
GraphSAGE with TransE protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.153
GraphSAGE with ProtTrans protein embeddings (sequence) (PPI embeddings study)Evaluation	0.138
MLP on concatenated protein embeddings (protein-only baseline, no network) with majority vote across the representations (PPI embeddings study)Evaluation	0.124
NGNN with GraphSAGE with ProstT5 protein embeddings (structure) (PPI embeddings study)Evaluation	0.118
MLP on concatenated protein embeddings (protein-only baseline, no network) with PiFold protein embeddings (structure) (PPI embeddings study)Evaluation	0.117
MLP on concatenated protein embeddings (protein-only baseline, no network) with randomly initialised node features (baseline) (PPI embeddings study)Evaluation	0.0916
MLP on concatenated protein embeddings (protein-only baseline, no network) with ESM-IF1 protein embeddings (structure) (PPI embeddings study)Evaluation	0.0755
MLP on concatenated protein embeddings (protein-only baseline, no network) with ESM-2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.0704
MLP on concatenated protein embeddings (protein-only baseline, no network) with ProGen2 protein embeddings (sequence) (PPI embeddings study)Evaluation	0.061
GCN with randomly initialised node features (baseline) (PPI embeddings study)Evaluation	0.0588
NGNN with GCN with randomly initialised node features (baseline) (PPI embeddings study)Evaluation	0.0558
MLP on concatenated protein embeddings (protein-only baseline, no network) with ProstT5 protein embeddings (structure) (PPI embeddings study)Evaluation	0.0521
MLP on concatenated protein embeddings (protein-only baseline, no network) with ProtTrans protein embeddings (sequence) (PPI embeddings study)Evaluation	0.0521
GraphSAGE with ProstT5 protein embeddings (structure) (PPI embeddings study)Evaluation	0.0385
GCN with TransE protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)Evaluation	0.0382
GraphSAGE with randomly initialised node features (baseline) (PPI embeddings study)Evaluation	0.0104

All 77 result rows with coverage, uncertainty and sources

Filter evaluations

Tested model or methodProtocolDatasetEvaluation setupMetricEvidence origin

Applied filters: All linked evaluations

Exact evaluated configurations and original reported results
Tested configuration	Protocol and dataset	Finding	Evidence and details
Configuration: GCN with ComplEx protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.245 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GCN / ComplEx on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GCN', column 'ComplEx'
Configuration: GCN with majority vote across the representations (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.21 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GCN / Ensemble on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GCN', column 'Ensemble'
Configuration: GCN with ESM-2 protein embeddings (sequence) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.16 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GCN / ESM-2 on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GCN', column 'ESM-2'
Configuration: GCN with ESM-IF1 protein embeddings (structure) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.185 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GCN / ESM-IF1 on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GCN', column 'ESM-IF1'
Configuration: GCN with PiFold protein embeddings (structure) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.189 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GCN / PiFold on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GCN', column 'PiFold'
Configuration: GCN with ProGen2 protein embeddings (sequence) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.218 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GCN / ProGen2 on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GCN', column 'ProGen2'
Configuration: GCN with ProstT5 protein embeddings (structure) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.162 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GCN / ProstT5 on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GCN', column 'ProstT5'
Configuration: GCN with ProtTrans protein embeddings (sequence) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.168 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GCN / ProtTrans on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GCN', column 'ProtTrans'
Configuration: GCN with randomly initialised node features (baseline) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.0588 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GCN / Random on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GCN', column 'Random'
Configuration: GCN with RDF2Vec protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.255 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GCN / RDF2Vec on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GCN', column 'RDF2Vec'
Configuration: GCN with TransE protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.0382 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GCN / TransE on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GCN', column 'TransE'
Configuration: GraphSAGE with ComplEx protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.276 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GraphSAGE / ComplEx on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GS', column 'ComplEx'
Configuration: GraphSAGE with majority vote across the representations (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.167 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GraphSAGE / Ensemble on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GS', column 'Ensemble'
Configuration: GraphSAGE with ESM-2 protein embeddings (sequence) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.242 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GraphSAGE / ESM-2 on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GS', column 'ESM-2'
Configuration: GraphSAGE with ESM-IF1 protein embeddings (structure) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.155 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GraphSAGE / ESM-IF1 on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GS', column 'ESM-IF1'
Configuration: GraphSAGE with PiFold protein embeddings (structure) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.209 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GraphSAGE / PiFold on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GS', column 'PiFold'
Configuration: GraphSAGE with ProGen2 protein embeddings (sequence) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.185 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GraphSAGE / ProGen2 on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GS', column 'ProGen2'
Configuration: GraphSAGE with ProstT5 protein embeddings (structure) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.0385 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GraphSAGE / ProstT5 on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GS', column 'ProstT5'
Configuration: GraphSAGE with ProtTrans protein embeddings (sequence) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.138 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GraphSAGE / ProtTrans on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GS', column 'ProtTrans'
Configuration: GraphSAGE with randomly initialised node features (baseline) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.0104 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GraphSAGE / Random on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GS', column 'Random'
Configuration: GraphSAGE with RDF2Vec protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.198 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GraphSAGE / RDF2Vec on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GS', column 'RDF2Vec'
Configuration: GraphSAGE with TransE protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.153 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source GraphSAGE / TransE on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'GS', column 'TransE'
Configuration: MLP on concatenated protein embeddings (protein-only baseline, no network) with ComplEx protein embeddings (function (knowledge-graph embedding)) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.197 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source MLP on concatenated protein embeddings (protein-only baseline, no network) / ComplEx on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'MLP', column 'ComplEx'
Configuration: MLP on concatenated protein embeddings (protein-only baseline, no network) with majority vote across the representations (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.124 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source MLP on concatenated protein embeddings (protein-only baseline, no network) / Ensemble on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'MLP', column 'Ensemble'
Configuration: MLP on concatenated protein embeddings (protein-only baseline, no network) with ESM-2 protein embeddings (sequence) (PPI embeddings study)	Protocol: OGBl-ppa PPI link prediction, Hits@100 (Table 5) Dataset: OGBl-ppa PPI graph (embedding study)	0.0704 Hits at 100 (OGBl-ppa PPI link prediction) fraction · higher Uncertainty: Not reported by the source: Only a table-wide maximum SD is printed (Table 5/6 footnote) Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The text's summary does not match the tables: it lists ProtT5 among structure embeddings while the tables print ProstT5; it reports a GraphSAGE with DistMult result (+38% Hits@100 on HuRI) although DistMult is not a column; and the stated gains (for example ESM-2 with GraphSAGE +6% Hits@100 on OGBl-ppa) cannot be reproduced from the printed table values (GS random 0.0104, ESM-2 0.2421). Values are kept as printed; comparisons should rely on the tables only after a reviewer has judged this. Random-feature baselines on OGBl-ppa (GCN 0.0588, GS 0.0104 Hits@100) look far below the public OGB leaderboard values the authors say they reproduced (Supplementary Table S2, not retrieved; reviewer did not verify the leaderboard values).The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Results and discussion, first paragraph; Tables 5 and 6 Methods, coverage and source MLP on concatenated protein embeddings (protein-only baseline, no network) / ESM-2 on OGBl-ppa (Hits@100) lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100 Aggregation: Not reported The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Table 5, block 'OGBl-ppa', row 'MLP', column 'ESM-2'

Source checking is not independent reproduction. Release 2026-10-22-1b7cf027e4f6.

Methods and evaluation design

Procedure, tasks and evaluated configurations

Recorded evaluations

Each evaluation records what was tested and under which conditions.

Explore all linked results

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

No execution recipe linked to this protocol. Recipe availability does not establish a completed evaluation.

Author-reported evaluations: 7
External evaluations: 70

Literature evidence is not a Rewire measurement. Executed but unpublished runs and private review status are not included.

Null control

Proposed control: requires review

Select a task-valid null control after reviewing inputs and metric

Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.

This is a suggested selection rule, not a validated method or a measured score.

Conventional reference

Proposed control: requires review

Select an upstream conventional reference after reviewing the full protocol

Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.

This is a suggested selection rule, not a validated method or a measured score.

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

Coverage is derived from release 2026-10-22-1b7cf027e4f6. Source citations describe the original records; they do not validate an unreviewed baseline proposal. No results have been generated by this audit.

Run instructions

No runnable recipe has been reviewed for this protocol. Dataset access, model requirements, licences and compute requirements must be checked against its sources before execution.

Strengths, limitations and unresolved questions

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.

0 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-10-22-1b7cf027e4f6
Property and statement	Original source and location	Review and provenance

No evidence rows match these filters. Choose another scope or clear the search.

How to read the evidence

Sources and history

Release 2026-10-22-1b7cf027e4f6 · Record review: source checked

1 source record and release history

The role of protein embeddings for protein-protein interaction prediction with graph neural networks · Original source · Briefings in Bioinformatics 2026, bbag313; PMC13280947 full-text XML

Download this release (gzip)

Technical metadata and extraction receipts

Stable ID: lit-interactions-20261011-ppiemb-protocol-ogbl-ppa-hits-100

areas: molecular-interactions; proteins-complexes
contexts: research
protocol: OGBl-ppa PPI link prediction; Hits@100
version: bbag313
metric: hits-at-100
metric direction: higher
unit: fraction
limitations: The authors ran every configuration; none of the embedding models is theirs.; No per-cell uncertainty or run count is printed; the footnote of Tables 5 and 6 gives only the table-wide maximum standard deviation (0.0711 OGBl-ppa, 0.0928 HuRI).; The MLP row is a protein-only binary classifier on concatenated embeddings, not a GNN.; The text's summary of gains does not match the printed values; see the source's evidence concern.
source locator: Table 5, block 'OGBl-ppa'

Related records

part of: Protein embeddings for GNN PPI prediction (OGBl-ppa and HuRI)
uses data: OGBl-ppa PPI graph (embedding study)
assessment: GCN / ComplEx on OGBl-ppa (Hits@100)
assessment: GCN / Ensemble on OGBl-ppa (Hits@100)
assessment: GCN / ESM-2 on OGBl-ppa (Hits@100)
assessment: GCN / ESM-IF1 on OGBl-ppa (Hits@100)
assessment: GCN / PiFold on OGBl-ppa (Hits@100)
assessment: GCN / ProGen2 on OGBl-ppa (Hits@100)
assessment: GCN / ProstT5 on OGBl-ppa (Hits@100)
assessment: GCN / ProtTrans on OGBl-ppa (Hits@100)
assessment: GCN / Random on OGBl-ppa (Hits@100)
assessment: GCN / RDF2Vec on OGBl-ppa (Hits@100)
assessment: GCN / TransE on OGBl-ppa (Hits@100)
assessment: GraphSAGE / ComplEx on OGBl-ppa (Hits@100)
assessment: GraphSAGE / Ensemble on OGBl-ppa (Hits@100)
assessment: GraphSAGE / ESM-2 on OGBl-ppa (Hits@100)
assessment: GraphSAGE / ESM-IF1 on OGBl-ppa (Hits@100)
assessment: GraphSAGE / PiFold on OGBl-ppa (Hits@100)
assessment: GraphSAGE / ProGen2 on OGBl-ppa (Hits@100)
assessment: GraphSAGE / ProstT5 on OGBl-ppa (Hits@100)
assessment: GraphSAGE / ProtTrans on OGBl-ppa (Hits@100)
assessment: GraphSAGE / Random on OGBl-ppa (Hits@100)
assessment: GraphSAGE / RDF2Vec on OGBl-ppa (Hits@100)
assessment: GraphSAGE / TransE on OGBl-ppa (Hits@100)
assessment: MLP on concatenated protein embeddings (protein-only baseline, no network) / ComplEx on OGBl-ppa (Hits@100)
assessment: MLP on concatenated protein embeddings (protein-only baseline, no network) / Ensemble on OGBl-ppa (Hits@100)
assessment: MLP on concatenated protein embeddings (protein-only baseline, no network) / ESM-2 on OGBl-ppa (Hits@100)
assessment: MLP on concatenated protein embeddings (protein-only baseline, no network) / ESM-IF1 on OGBl-ppa (Hits@100)
assessment: MLP on concatenated protein embeddings (protein-only baseline, no network) / PiFold on OGBl-ppa (Hits@100)
assessment: MLP on concatenated protein embeddings (protein-only baseline, no network) / ProGen2 on OGBl-ppa (Hits@100)
assessment: MLP on concatenated protein embeddings (protein-only baseline, no network) / ProstT5 on OGBl-ppa (Hits@100)
assessment: MLP on concatenated protein embeddings (protein-only baseline, no network) / ProtTrans on OGBl-ppa (Hits@100)
assessment: MLP on concatenated protein embeddings (protein-only baseline, no network) / Random on OGBl-ppa (Hits@100)
assessment: MLP on concatenated protein embeddings (protein-only baseline, no network) / RDF2Vec on OGBl-ppa (Hits@100)
assessment: MLP on concatenated protein embeddings (protein-only baseline, no network) / TransE on OGBl-ppa (Hits@100)
assessment: NGNN with GCN / ComplEx on OGBl-ppa (Hits@100)
assessment: NGNN with GCN / Ensemble on OGBl-ppa (Hits@100)
assessment: NGNN with GCN / ESM-2 on OGBl-ppa (Hits@100)
assessment: NGNN with GCN / ESM-IF1 on OGBl-ppa (Hits@100)
assessment: NGNN with GCN / PiFold on OGBl-ppa (Hits@100)
assessment: NGNN with GCN combined with SEAL / ComplEx on OGBl-ppa (Hits@100)
assessment: NGNN with GCN combined with SEAL / Ensemble on OGBl-ppa (Hits@100)
assessment: NGNN with GCN combined with SEAL / ESM-2 on OGBl-ppa (Hits@100)
assessment: NGNN with GCN combined with SEAL / ESM-IF1 on OGBl-ppa (Hits@100)
assessment: NGNN with GCN combined with SEAL / PiFold on OGBl-ppa (Hits@100)
assessment: NGNN with GCN combined with SEAL / ProGen2 on OGBl-ppa (Hits@100)
assessment: NGNN with GCN combined with SEAL / ProstT5 on OGBl-ppa (Hits@100)
assessment: NGNN with GCN combined with SEAL / ProtTrans on OGBl-ppa (Hits@100)
assessment: NGNN with GCN combined with SEAL / Random on OGBl-ppa (Hits@100)
assessment: NGNN with GCN combined with SEAL / RDF2Vec on OGBl-ppa (Hits@100)
assessment: NGNN with GCN combined with SEAL / TransE on OGBl-ppa (Hits@100)
assessment: NGNN with GCN / ProGen2 on OGBl-ppa (Hits@100)
assessment: NGNN with GCN / ProstT5 on OGBl-ppa (Hits@100)
assessment: NGNN with GCN / ProtTrans on OGBl-ppa (Hits@100)
assessment: NGNN with GCN / Random on OGBl-ppa (Hits@100)
assessment: NGNN with GCN / RDF2Vec on OGBl-ppa (Hits@100)
assessment: NGNN with GCN / TransE on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE / ComplEx on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE / Ensemble on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE / ESM-2 on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE / ESM-IF1 on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE / PiFold on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE combined with SEAL / ComplEx on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE combined with SEAL / Ensemble on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE combined with SEAL / ESM-2 on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE combined with SEAL / ESM-IF1 on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE combined with SEAL / PiFold on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE combined with SEAL / ProGen2 on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE combined with SEAL / ProstT5 on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE combined with SEAL / ProtTrans on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE combined with SEAL / Random on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE combined with SEAL / RDF2Vec on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE combined with SEAL / TransE on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE / ProGen2 on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE / ProstT5 on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE / ProtTrans on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE / Random on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE / RDF2Vec on OGBl-ppa (Hits@100)
assessment: NGNN with GraphSAGE / TransE on OGBl-ppa (Hits@100)

Suggest a correction