OGB
ogbl-ddi
Structure
1Graphs
4,267Nodes
1,334,889Edges
Coverage
135Results
1Variants
22Papers
- NeurIPS 2025: Bridging Theory and Practice in Link Representation with Graph Neural Networks Bridging Theory and Practice in Link Representation with Graph Neural Networks NeurIPS 2025
- Using Random Noise Equivariantly to Boost Graph Neural Networks Universally Using Random Noise Equivariantly to Boost Graph Neural Networks Universally
- NeurIPS 2024: PageRank Bandits for Link Prediction PageRank Bandits for Link Prediction NeurIPS 2024
- Link Prediction with Untrained Message Passing Layers Link Prediction with Untrained Message Passing Layers
- OpenGraph: Towards Open Graph Foundation Models OpenGraph: Towards Open Graph Foundation Models
- Ensemble Learning for Graph Neural Networks Ensemble Learning for Graph Neural Networks
- Efficient Link Prediction via GNN Layers Induced by Negative Sampling Efficient Link Prediction via GNN Layers Induced by Negative Sampling
- ICLR 2023: Revisiting Link Prediction: a data perspective Revisiting Link Prediction: a data perspective ICLR 2023
- JMLR 2023: Improving Graph Neural Networks on Multi-node Tasks with the Labeling Trick Improving Graph Neural Networks on Multi-node Tasks with the Labeling Trick JMLR 2023
- NeurIPS 2023: Graph Mixture of Experts: Learning on Large-Scale Graphs with Explicit Diversity Modeling Graph Mixture of Experts: Learning on Large-Scale Graphs with Explicit Diversity Modeling NeurIPS 2023
- Simplifying Subgraph Representation Learning for Scalable Link Prediction Simplifying Subgraph Representation Learning for Scalable Link Prediction
- Bring Your Own View: Graph Neural Networks for Link Prediction with Personalized Subgraph Selection Bring Your Own View: Graph Neural Networks for Link Prediction with Personalized Subgraph Selection
- ICLR 2022: Graph Neural Networks for Link Prediction with Subgraph Sketching Graph Neural Networks for Link Prediction with Subgraph Sketching ICLR 2022
- NeurIPS 2022: Neo-GNNs: Neighborhood Overlap-aware Graph Neural Networks for Link Prediction Neo-GNNs: Neighborhood Overlap-aware Graph Neural Networks for Link Prediction NeurIPS 2022
- ICLR 2022: Equivariant and Stable Positional Encoding for More Powerful Graph Neural Networks Equivariant and Stable Positional Encoding for More Powerful Graph Neural Networks ICLR 2022
- MGAE: Masked Autoencoders for Self-Supervised Learning on Graphs MGAE: Masked Autoencoders for Self-Supervised Learning on Graphs
- Pairwise Learning for Neural Link Prediction Pairwise Learning for Neural Link Prediction
- ICML 2021: Learning from Counterfactual Links for Link Prediction Learning from Counterfactual Links for Link Prediction ICML 2021
- Adaptive Graph Diffusion Networks Adaptive Graph Diffusion Networks
- NeurIPS 2020: Labeling Trick: A Theory of Using Graph Neural Networks for Multi-Node Representation Learning Labeling Trick: A Theory of Using Graph Neural Networks for Multi-Node Representation Learning NeurIPS 2020
- Formatting Instructions for ICLR 2021 Conference Submissions Formatting Instructions for ICLR 2021 Conference Submissions
- NeurIPS 2020: OGB OGB NeurIPS 2020
Link prediction Rank true drug-drug interaction edges above sampled non-interacting pairs. The standard graph supplies topology but no intrinsic drug node features or typed interaction labels. Evaluated by Hits@20.
Dataset details
Task
Rank true drug-drug interaction edges above sampled non-interacting pairs. OGB reports Hits@20 with each positive ranked among approximately 100,000 random negatives.
Data
The homogeneous, undirected, unweighted DrugBank graph contains 4,267 drugs. An edge means the combined effect of the drug pair differs substantially from the effect expected if the drugs acted independently.
Features
The standard graph supplies topology but no intrinsic drug node features or typed interaction labels. Models using molecular structures, targets, or descriptions are adding external data.
Splits and evaluation
OGB groups edges using drug protein targets so test drugs predominantly act on proteins different from those associated with train and validation drugs. This tests transfer across mechanisms of action rather than a random edge holdout.
Quirks and pitfalls
No edge is not the same as a clinically verified safe combination, and DrugBank includes both approved and experimental compounds. The graph collapses interaction types and direction. External chemical features can radically change the task and must be reported.
Sources
SOTA Progression
new SOTA
matches SOTA at that time
other proposed result
other baseline result
Leaderboard
| # | Model | Hits@20 ↑ | MRR | Recall@20 | Paper |
|---|---|---|---|---|---|
| 1 | ELGNN | 0.9777
±
0.0037 | — | — | |
| 2 | GIDN | 0.9542
±
0.0000 | — | — | |
| 3 | AGDN | 0.9538
±
0.0094 | — | — | |
| 4 | PSG | 0.9284
±
0.0047 | — | — | |
| 5 | PLNLP | 0.9088
±
0.0313 | — | — | |
| 6 | SUREL | 0.8974
±
0.0018 | — | — | |
| 7 | PRB | 0.8831
±
0.0436 | — | — | |
| 8 | SAGE+Edge Attr | 0.8781
±
0.0447 | — | — | |
| 9 | GraphSAGE + Edge Attr | 0.8781
±
0.0474 | — | — | |
| 10 | CFLP (w/ JKNet) | 0.8608
±
0.0198 | — | — | |
| 11 | CFLP w/ JKNet | 0.8608
±
0.0198 | — | — | |
| 12 | ELPH | 0.8319
±
0.0212 | — | — | |
| 13 | NCN | 0.8232
±
0.0610 | — | — | |
| 14 | YinYanGNN | 0.8092
±
0.0335 | — | — | |
| 15 | BUDDY | 0.7851
±
0.0136 | — | — | |
| 16 | ENGNN | 0.7761
±
0.0450 | — | — | |
| 17 | CFLP w/ GSAGE | 0.7549
±
0.0433 | — | — | |
| 18 | SAGE+Proposal Set | 0.7495
±
0.0317 | — | — | |
| 19 | NCNC | 0.7023
±
0.1211 | — | — | |
| 20 | SAGE | 0.6719
±
0.0118 | — | — | |
| 21 | GAT | 0.6683
±
0.0223 | — | — | |
| 22 | MGAE-SAGE | 0.6600
±
0.0949 | — | — | |
| 23 | MGAE-GCN | 0.6591
±
0.0350 | — | — | |
| 24 | NCN | 0.6582
±
0.0266 | — | — | |
| 25 | GCN | 0.6476
±
0.0145 | — | — | |
| 26 | Node2Vec | 0.6363
±
0.0205 | — | — | |
| 27 | Neo-GNN | 0.6357
±
0.0352 | — | — | |
| 28 | GCN+LRGA | 0.6230
±
0.0912 | — | — | |
| 29 | LRGA+GCN | 0.6230
±
0.0910 | — | — | |
| 30 | GCN+JKNet | 0.6056
±
0.0869 | — | — | |
| 31 | JKNet | 0.6056
±
0.0869 | — | — | |
| 32 | MF | 0.5950
±
0.0168 | — | — | |
| 33 | BUDDY | 0.5871
±
0.0163 | — | — | |
| 34 | JK-Net | 0.5798
±
0.0688 | — | — | |
| 35 | GraphSage-PS2 | 0.5690
±
0.0532 | — | — | |
| 36 | GraphConv + GAE | 0.5636
±
0.0045 | — | — | |
| 37 | PE-SAGE-DW | 0.5616
±
0.0550 | — | — | |
| 38 | SNCNC (SGIN) | 0.5453
±
0.1102 | — | — | |
| 39 | GIN + GAE | 0.5421
±
0.0095 | — | — | |
| 40 | SAGE | 0.5390
±
0.0474 | — | — | |
| 41 | GraphSAGE | 0.5390
±
0.0474 | — | — | |
| 42 | GSAGE | 0.5390
±
0.0474 | — | — | |
| 43 | CFLP w/ GCN | 0.5251
±
0.0109 | — | — | |
| 44 | PE-SAGE-LE | 0.5149
±
0.0971 | — | — | |
| 45 | PEG | 0.5012
±
0.0655 | — | — | |
| 46 | SEAL | 0.4974
±
0.0239 | — | — | |
| 47 | GAE-PS2 | 0.4953
±
0.0599 | — | — | |
| 48 | PEG-DW | 0.4793
±
0.0021 | — | — | |
| 49 | NCNC (GCN) | 0.4653
±
0.2924 | — | — | |
| 50 | GCN | 0.4460
±
0.0887 | — | — |
SOTA performance
set a new SOTA at the time
matched the SOTA at the time