MoleculeNet
Tox21
Structure
7,831Graphs
19Nodes
19Edges
12Classes
9Node Feature Dim
3Edge Feature Dim
Coverage
457Results
1Variants
65Papers
- A Systematic Survey and Benchmark of Deep Learning for Molecular Property Prediction in the Foundation Model Era A Systematic Survey and Benchmark of Deep Learning for Molecular Property Prediction in the Foundation Model Era
- <SO$ _k$>: One LLM Token for Explicit Graph Structural Understanding <SO$ _k$>: One LLM Token for Explicit Graph Structural Understanding
- Multi-scale Graph Autoregressive Modeling: Molecular Property Prediction via Next Token Prediction Multi-scale Graph Autoregressive Modeling: Molecular Property Prediction via Next Token Prediction
- On The Expressive Power of GNN Derivatives On The Expressive Power of GNN Derivatives
- A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery
- WWW 2025: Graph Self-Supervised Learning with Learnable Structural and Positional Encodings Graph Self-Supervised Learning with Learnable Structural and Positional Encodings WWW 2025
- ICML 2025: Graph Generative Pre-trained Transformer Graph Generative Pre-trained Transformer ICML 2025
- ICML 2024: Towards Graph Foundation Models: Learning Generalities Across Graphs via Task-Trees Towards Graph Foundation Models: Learning Generalities Across Graphs via Task-Trees ICML 2024
- AAAI 2024: Pre-training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck Pre-training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck AAAI 2024
- TMLR 2024: Towards Graph Foundation Models: A Study on the Generalization of Positional and Structural Encodings Towards Graph Foundation Models: A Study on the Generalization of Positional and Structural Encodings TMLR 2024
- SpeGCL: Self-supervised Graph Spectrum Contrastive Learning without Positive Samples SpeGCL: Self-supervised Graph Spectrum Contrastive Learning without Positive Samples
- NeurIPS 2024: DiGRAF: Diffeomorphic Graph-Adaptive Activation Function DiGRAF: Diffeomorphic Graph-Adaptive Activation Function NeurIPS 2024
- Scalable Expressiveness through Preprocessed Graph Perturbations Scalable Expressiveness through Preprocessed Graph Perturbations
- NeurIPS 2024: Text-space Graph Foundation Models: Comprehensive Benchmarks and New Insights Text-space Graph Foundation Models: Comprehensive Benchmarks and New Insights NeurIPS 2024
- ICML 2024: UniCorn: A Unified Contrastive Learning Approach for Multi-view Molecular Representation Learning UniCorn: A Unified Contrastive Learning Approach for Multi-view Molecular Representation Learning ICML 2024
- NeurIPS 2024: SubGDiff: A Subgraph Diffusion Model to Improve Molecular Representation Learning SubGDiff: A Subgraph Diffusion Model to Improve Molecular Representation Learning NeurIPS 2024
- ICML 2024: Community-Invariant Graph Contrastive Learning Community-Invariant Graph Contrastive Learning ICML 2024
- ICLR 2024: Decoupling Weighing and Selecting for Integrating Multiple Graph Pre-training Tasks Decoupling Weighing and Selecting for Integrating Multiple Graph Pre-training Tasks ICLR 2024
- Subgraph-level Universal Prompt Tuning Subgraph-level Universal Prompt Tuning
- UGMAE: A Unified Framework for Graph Masked Autoencoders UGMAE: A Unified Framework for Graph Masked Autoencoders
- Impact of Domain Knowledge and Multi-Modality on Intelligent Molecular Property Prediction: A Systematic Survey Impact of Domain Knowledge and Multi-Modality on Intelligent Molecular Property Prediction: A Systematic Survey
- AAAI 2023: Fine-tuning Graph Neural Networks by Preserving Graph Generative Patterns Fine-tuning Graph Neural Networks by Preserving Graph Generative Patterns AAAI 2023
- Permutation-Invariant graph partitioning: How graph neural networks capture structural interactions? Permutation-Invariant graph partitioning: How graph neural networks capture structural interactions?
- NeurIPS 2023: Graph Convolutions Enrich the Self-Attention in Transformers! Graph Convolutions Enrich the Self-Attention in Transformers! NeurIPS 2023
- NeurIPS 2023: Better with Less: A Data-Active Perspective on Pre-Training Graph Neural Networks Better with Less: A Data-Active Perspective on Pre-Training Graph Neural Networks NeurIPS 2023
- ICLR 2023: Efficient Subgraph GNNs by Learning Effective Selection Policies Efficient Subgraph GNNs by Learning Effective Selection Policies ICLR 2023
- NeurIPS 2023: Rethinking Tokenizer and Decoder in Masked Graph Modeling for Molecules Rethinking Tokenizer and Decoder in Masked Graph Modeling for Molecules NeurIPS 2023
- UniMAP: Universal SMILES-Graph Representation Learning UniMAP: Universal SMILES-Graph Representation Learning
- Atom-Motif Contrastive Transformer for Molecular Property Prediction Atom-Motif Contrastive Transformer for Molecular Property Prediction
- Deep Prompt Tuning for Graph Transformers Deep Prompt Tuning for Graph Transformers
- Motif-aware Attribute Masking for Molecular Graph Pre-training Motif-aware Attribute Masking for Molecular Graph Pre-training
- AAAI 2023: 3D Denoisers are Good 2D Teachers: Molecular Pretraining via Denoising and Cross-Modal Distillation 3D Denoisers are Good 2D Teachers: Molecular Pretraining via Denoising and Cross-Modal Distillation AAAI 2023
- Graph Self-Contrast Representation Learning Graph Self-Contrast Representation Learning
- P-tensors: a General Framework for Higher Order Message Passing in Subgraph Neural Networks P-tensors: a General Framework for Higher Order Message Passing in Subgraph Neural Networks
- KDD 2023: Improving Expressivity of GNNs with Subgraph-specific Factor Embedded Normalization Improving Expressivity of GNNs with Subgraph-specific Factor Embedded Normalization KDD 2023
- ICML 2023: A Group Symmetric Stochastic Differential Equation Model for Molecule Multi-modal Pretraining A Group Symmetric Stochastic Differential Equation Model for Molecule Multi-modal Pretraining ICML 2023
- gimlet: A Unified Graph-Text Model for Instruction-Based Molecule Zero-Shot Learning gimlet: A Unified Graph-Text Model for Instruction-Based Molecule Zero-Shot Learning
- Robust Ante-hoc Graph Explainer using Bilevel Optimization Robust Ante-hoc Graph Explainer using Bilevel Optimization
- AAAI 2023: Union Subgraph Neural Networks Union Subgraph Neural Networks AAAI 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
- ICML 2022: A Generalization of ViT/MLP-Mixer to Graphs A Generalization of ViT/MLP-Mixer to Graphs ICML 2022
- NeurIPS 2022: Universal Prompt Tuning for Graph Neural Networks Universal Prompt Tuning for Graph Neural Networks NeurIPS 2022
- Improving Molecular Pretraining with Complementary Featurizations Improving Molecular Pretraining with Complementary Featurizations
- Motif-based Graph Representation Learning with Application to Chemical Molecules Motif-based Graph Representation Learning with Application to Chemical Molecules
- KDD 2022: Unified 2D and 3D Pre-Training of Molecular Representations Unified 2D and 3D Pre-Training of Molecular Representations KDD 2022
- NeurIPS 2022: Does GNN Pretraining Help Molecular Representation? Does GNN Pretraining Help Molecular Representation? NeurIPS 2022
- KDD 2022: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction KDD 2022
- ICML 2022: Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning ICML 2022
- KDD 2022: GraphMAE: Self-Supervised Masked Graph Autoencoders GraphMAE: Self-Supervised Masked Graph Autoencoders KDD 2022
- Improving Molecular Contrastive Learning via Faulty Negative Mitigation and Decomposed Fragment Contrast Improving Molecular Contrastive Learning via Faulty Negative Mitigation and Decomposed Fragment Contrast
- NeurIPS 2022: Graph Self-supervised Learning with Accurate Discrepancy Learning Graph Self-supervised Learning with Accurate Discrepancy Learning NeurIPS 2022
- OOD-GNN: Out-of-Distribution Generalized Graph Neural Network OOD-GNN: Out-of-Distribution Generalized Graph Neural Network
- ICLR 2021: Graph Neural Networks with Learnable Structural and Positional Representations Graph Neural Networks with Learnable Structural and Positional Representations ICLR 2021
- ICLR 2021: 16 Pre-training Molecular Graph Representation with 3D Geometry 16 Pre-training Molecular Graph Representation with 3D Geometry ICLR 2021
- NeurIPS 2021: Motif-based Graph Self-Supervised Learning for Molecular Property Prediction Motif-based Graph Self-Supervised Learning for Molecular Property Prediction NeurIPS 2021
- NeurIPS 2021: Reconstruction for Powerful Graph Representations Reconstruction for Powerful Graph Representations NeurIPS 2021
- AAAI 2021: GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction AAAI 2021
- ChemRL-GEM: Geometry Enhanced Molecular Representation Learning for Property Prediction ChemRL-GEM: Geometry Enhanced Molecular Representation Learning for Property Prediction
- ICLR 2021: Accurate Learning of Graph Representations with Graph Multiset Pooling Accurate Learning of Graph Representations with Graph Multiset Pooling ICLR 2021
- Graph convolutions that can finally model local structure Graph convolutions that can finally model local structure
- AAAI 2020: Rethinking Graph Regularization for Graph Neural Networks Rethinking Graph Regularization for Graph Neural Networks AAAI 2020
- ICLR 2020: Memory-Based Graph Networks Memory-Based Graph Networks ICLR 2020
- ICLR 2019: Strategies for Pre-training Graph Neural Networks Strategies for Pre-training Graph Neural Networks ICLR 2019
- Towards Interpretable Sparse Graph Representation Learning with Laplacian Pooling Towards Interpretable Sparse Graph Representation Learning with Laplacian Pooling
- AAAI 2018: Adaptive Graph Convolutional Neural Networks Adaptive Graph Convolutional Neural Networks AAAI 2018
12-task multi-label graph classification Multi-task toxicity prediction (12 tasks) on NCI Tox21 challenge compounds. SMILES and labels are fixed; molecular graph featurization is implementation-defined. Evaluated by ROC-AUC.
Dataset details
Task
Multi-task toxicity prediction (12 tasks) on NCI Tox21 challenge compounds. The primary catalog metric is ROC-AUC.
Data
Multi-task toxicity prediction (12 tasks) on NCI Tox21 challenge compounds.
Features
MoleculeNet distributes SMILES and targets, not one mandatory graph featurization. A graph implementation normally derives atom and bond fields with RDKit or DeepChem; results depend on featurizer, salt handling, stereochemistry, and invalid-molecule policy.
Splits and evaluation
The catalog records these protocols or variants: Scaffold split. Evaluation uses ROC-AUC (higher is better). Exact masks or folds must come from the cited release.
Quirks and pitfalls
Small samples, imbalance, missing assay labels, duplicate compounds, and scaffold leakage are common. Missing labels are not negatives. Always report split algorithm, featurizer, metric aggregation, and dataset version; random and scaffold results are not comparable.
Sources
SOTA Progression
new SOTA
matches SOTA at that time
other proposed result
other baseline result
Leaderboard
| # | Model | ROC-AUC ↑ | Accuracy | ROC | ROCAUC | Paper |
|---|---|---|---|---|---|---|
| 1 | HIMP | 0.8740
±
0.0050 Accuracy
0.8740
±
0.0050 | 0.8740
±
0.0050 | — | — | |
| 2 | GINE w/ VN | 0.8720
±
0.0030 Accuracy
0.8720
±
0.0030 | 0.8720
±
0.0030 | — | — | |
| 3 | NaiveGINe+ K=3 w/ VN | 0.8700
±
0.0040 Accuracy
0.8700
±
0.0040 | 0.8700
±
0.0040 | — | — | |
| 4 | GINe+ K=1 w/ VN | 0.8670
±
0.0040 Accuracy
0.8670
±
0.0040 | 0.8670
±
0.0040 | — | — | |
| 5 | GINE | 0.8668
±
0.0077 | — | — | — | |
| 6 | GCN w/ VN | 0.8590
±
0.0050 Accuracy
0.8590
±
0.0050 | 0.8590
±
0.0050 | — | — | |
| 7 | GINE | 0.8500
±
0.0090 Accuracy
0.8500
±
0.0090 | 0.8500
±
0.0090 | — | — | |
| 8 | GeomGCL | 0.8500 | — | — | — | |
| 9 | MPNNs | 0.8495
±
0.0058 | — | — | — | |
| 10 | KPGT | 0.8480
±
0.0130 | — | — | — | |
| 11 | DMPNN | 0.8480 | — | — | — | |
| 12 | LiGhT FT | 0.8440
±
0.0040 | — | — | — | |
| 13 | LiGhT DeepGPT | 0.8430
±
0.0040 | — | — | — | |
| 14 | GCN | 0.8400
±
0.0040 Accuracy
0.8400
±
0.0040 | 0.8400
±
0.0040 | — | — | |
| 15 | Contextpred+Sup | 0.8400
±
0.0230 | — | — | — | |
| 16 | GROVER | 0.8380
±
0.0170 | — | — | — | |
| 17 | GeomMPNN | 0.8380 | — | — | — | |
| 18 | CoMPT | 0.8360 | — | — | — | |
| 19 | AttentiveFP | 0.8350 | — | — | — | |
| 20 | lapool | 0.8342
±
0.0097 | — | — | — | |
| 21 | MAT | 0.8340 | — | — | — | |
| 22 | GraphGPS FT | 0.8320
±
0.0090 | — | — | — | |
| 23 | GraphGPS DeepGPT | 0.8320
±
0.0120 | — | — | — | |
| 24 | DeeperGCN | 0.8310
±
0.0040 Accuracy
0.8310
±
0.0040 | 0.8310
±
0.0040 | — | — | |
| 25 | GCN | 0.8290
±
0.0060 | — | — | — | |
| 26 | GIN | 0.8290
±
0.0069 | — | — | — | |
| 27 | Masking+Sup | 0.8280
±
0.0130 | — | — | — | |
| 28 | MemGNN | 0.8280
±
0.0040 | — | — | — | |
| 29 | CamS-LLaMA | 0.8270 | — | — | — | |
| 30 | OFA | 0.8250 | — | — | — | |
| 31 | MoCL | 0.8240 | — | — | — | |
| 32 | DiffPool | 0.8237
±
0.0090 | — | — | — | |
| 33 | JOAO | 0.8230
±
0.0220 | — | — | — | |
| 34 | Kernel SVM | 0.8220
±
0.0060 | — | — | — | |
| 35 | Weave | 0.8200
±
0.0100 | — | — | — | |
| 36 | Masking Node + Supervised | 0.8193
±
0.0179 | — | — | — | |
| 37 | EdgePred | 0.8180
±
0.0250 | — | — | — | |
| 38 | Context Prediction + Supervised | 0.8171
±
0.0179 | — | — | — | |
| 39 | Infomax | 0.8160
±
0.0210 | — | — | — | |
| 40 | Graph U-Net | 0.8141
±
0.0060 | — | — | — | |
| 41 | Graphormer DeepGPT | 0.8130
±
0.0100 | — | — | — | |
| 42 | Supervised | 0.8113
±
0.0239 | — | — | — | |
| 43 | Bypass | 0.8100
±
0.0130 | — | — | — | |
| 44 | S-CGIB | 0.8094
±
0.0017 | — | — | — | |
| 45 | GIN | 0.8094
±
0.0017 | — | — | — | |
| 46 | EdgePred SUPT_soft | 0.8090 | — | — | — | |
| 47 | EdgePred SUPT_hard | 0.8076 | — | — | — | |
| 48 | GatedGCN | 0.8070
±
0.0110 | — | — | — | |
| 49 | ContextPred | 0.8060
±
0.0120 | — | — | — | |
| 50 | GraphCL | 0.8050
±
0.0170 | — | — | — |
SOTA performance
set a new SOTA at the time
matched the SOTA at the time