Community-Invariant Graph Contrastive Learning

Shiyin Tan, Dongyuan Li, Renhe Jiang, Ying Zhang, Manabu Okumura

Abstract

Graph augmentation has received great attention in recent years for graph contrastive learning (GCL) to learn well-generalized node/graph representations. However, mainstream GCL methods often favor randomly disrupting graphs for augmentation, which shows limited generalization and inevitably leads to the corruption of high-level graph information, i.e., the graph community. Moreover, current knowledge-based graph augmentation methods can only focus on either topology or node features, causing the model to lack robustness against various types of noise. To address these limitations, this research investigated the role of the graph community in graph augmentation and figured out its crucial advantage for learnable graph augmentation. Based on our observations, we propose a community-invariant GCL framework to maintain graph community structure during learnable graph augmentation. By maximizing the spectral changes, this framework unifies the constraints of both topology and feature augmentation, enhancing the model's robustness. Empirical evidence on 21 benchmark datasets demonstrates the exclusive merits of our framework. Code is released on Github (https://github.com/ShiyinTan/CI-GCL.git).

Benchmark Results

Tested on 12 datasets 18 baseline values reported too low

Tables show presented results in global context (best 3 papers + GCN + MLP).

TU Dortmund

MUTAG

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	MSH-GNN GNN global	0.9910 ±0.0030	Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing
2	ID-GNN+BEC Hyb global	0.9830 ±0.0360	Depth-Adaptive Graph Neural Networks via Learnable Bakry-Émery Curvature
3	SupCosine Hyb global	0.9830 ±0.0250	Supervised Contrastive Learning with Structure Inference for Graph Classification
39	GIN GNN global	0.9330 ±0.0290	ProtGNN: Towards Self-Explaining Graph Neural Networks
155	GCS GNN baseline actual #155 vs paper #326 171 places worse	0.9045 ±0.0081 best reported	actual #155 vs paper #326 171 places worse paper 0.8819 ±0.0090 best today: SpeGCL: Self-supervised Graph Spectrum Contrastive Learning without Positive Samples (2024) paper 0.8819 ±0.0090 Δ today 0.0226
178	AD-GCL GNN baseline actual #178 vs paper #213 35 places worse	0.9028 ±0.0401 best reported	actual #178 vs paper #213 35 places worse paper 0.8970 ±0.0100 best at pub: Unifying Graph Contrastive Learning via Graph Message Augmentation (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.8970 ±0.0100 Δ at pub 0.0058
179	SEGA GNN baseline	0.9021 ±0.0070	This paper
209	CI+AD-GCL Hyb baseline	0.8988 ±0.0070	This paper
223	CI-GCL GNN ★ proposed	0.8967 ±0.0090	This paper
257	InfoGraph GNN baseline actual #257 vs paper #272 15 places worse	0.8930 ±0.0514 best reported	actual #257 vs paper #272 15 places worse paper 0.8901 ±0.0110 best at pub: Improving Expressivity of GNNs with Subgraph-specific Factor Embedded Normalization (KDD 2023) Difference within reported uncertainty; not counted as underperformance. paper 0.8901 ±0.0110 Δ at pub 0.0029
302	AutoGCL GNN baseline actual #302 vs paper #446 144 places worse	0.8864 ±0.0108 best reported	actual #302 vs paper #446 144 places worse paper 0.8515 ±0.0110 best at pub: GIMM: InfoMin-Max for Automated Graph Contrastive Learning (2023) best today: SpeGCL: Self-supervised Graph Spectrum Contrastive Learning without Positive Samples (2024) paper 0.8515 ±0.0110 Δ at pub 0.0349 Δ today 0.0349
351	JOAO GNN baseline actual #351 vs paper #370 19 places worse	0.8770 ±0.0080 best reported	actual #351 vs paper #370 19 places worse paper 0.8735 ±0.0100 best at pub: Graph Self-Supervised Learning: A Survey (2021) best today: Learning Graph Augmentations to Learn Graph Representations (2022) Difference within reported uncertainty; not counted as underperformance. paper 0.8735 ±0.0100 Δ at pub 0.0035 Δ today 0.0035
400	CI+AutoGCL Hyb baseline	0.8673 ±0.0100	This paper
403	MLP code ↗ Trad global	0.8660 ±0.0495	Breaking the Limits of Message Passing Graph Neural Networks code ↗
451	GCL-SPAN GNN baseline	0.8500 ±0.0080	This paper

NCI1

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	WL-MLP GT global	0.9954	1-WL Expressiveness Is (Almost) All You Need
2	GIN GNN global	0.9934	1-WL Expressiveness Is (Almost) All You Need
3	GCN GNN global	0.9727	1-WL Expressiveness Is (Almost) All You Need
104	GraphCL GNN baseline actual #104 vs paper #325 221 places worse	0.8250 ±0.0013 best reported	actual #104 vs paper #325 221 places worse paper 0.7787 ±0.0040 best at pub: Label-invariant Augmentation for Semi-Supervised Graph Classification (NeurIPS 2022) paper 0.7787 ±0.0040 Δ at pub 0.0463
133	AutoGCL GNN baseline actual #133 vs paper #304 171 places worse	0.8200 ±0.0029 best reported	actual #133 vs paper #304 171 places worse paper 0.7832 ±0.0050 best at pub: Unifying Graph Contrastive Learning via Graph Message Augmentation (2024) best today: SpeGCL: Self-supervised Graph Spectrum Contrastive Learning without Positive Samples (2024) paper 0.7832 ±0.0050 Δ at pub 0.0368 Δ today 0.0368
131	GraphSAGE code ↗ GNN global	0.8200 ±0.0100	Weisfeiler and Leman Go Infinite: Spectral and Combinatorial Pre-Colorings code ↗
195	CI-GCL GNN ★ proposed	0.8050 ±0.0050	This paper
264	SEGA GNN baseline	0.7900 ±0.0070	This paper
296	CI+AutoGCL Hyb baseline	0.7847 ±0.0070	This paper
301	JOAO GNN baseline actual #301 vs paper #317 16 places worse	0.7840 ±0.0050 best reported	actual #301 vs paper #317 16 places worse paper 0.7807 ±0.0040 best at pub: Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning (ICML 2022) Difference within reported uncertainty; not counted as underperformance. paper 0.7807 ±0.0040 Δ at pub 0.0033
338	InfoGraph GNN baseline actual #338 vs paper #386 48 places worse	0.7750 ±0.0074 best reported	actual #338 vs paper #386 48 places worse paper 0.7620 ±0.0100 best at pub: Graph Masked Autoencoders with Transformers (2022) Difference within reported uncertainty; not counted as underperformance. paper 0.7620 ±0.0100 Δ at pub 0.0130
344	GCS GNN baseline actual #344 vs paper #353 9 places worse	0.7737 ±0.0030 best reported	actual #344 vs paper #353 9 places worse paper 0.7718 ±0.0030 best today: SpeGCL: Self-supervised Graph Spectrum Contrastive Learning without Positive Samples (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.7718 ±0.0030 Δ today 0.0019
356	MVGRL GNN baseline actual #356 vs paper #371 15 places worse	0.7700 ±0.0080 best reported	actual #356 vs paper #371 15 places worse paper 0.7664 ±0.0030 best at pub: InfoGCL: Information-Aware Graph Contrastive Learning (NeurIPS 2021) best today: Graph Self-Supervised Learning with Learnable Structural and Positional Encodings (WWW 2025) Difference within reported uncertainty; not counted as underperformance. paper 0.7664 ±0.0030 Δ at pub 0.0036 Δ today 0.0036
405	GCL-SPAN GNN baseline	0.7543 ±0.0040	This paper
427	CI+AD-GCL Hyb baseline	0.7435 ±0.0050	This paper

PROTEINS

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	MSH-GNN GNN global	0.9410 ±0.0330	Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing
2	GRDL GNN global	0.8260 ±0.0120	Graph Classification via Reference Distribution Learning: Theory and Practice
3	GMN code ↗ Hyb global	0.8225	Memory-Based Graph Networks code ↗
32	GIN GNN global	0.7880 ±0.0410	Permutation-Invariant graph partitioning: How graph neural networks capture structural …
107	CI-GCL GNN ★ proposed	0.7650 ±0.0010	This paper
146	SEGA GNN baseline	0.7601 ±0.0040	This paper
149	MLP Trad global	0.7600 ±0.0320	Simplifying Graph Convolutional Networks with Redundancy-Free Neighbors
169	GCL-SPAN GNN baseline	0.7578 ±0.0040	This paper
245	GCS GNN baseline actual #245 vs paper #309 64 places worse	0.7502 ±0.0081 best reported	actual #245 vs paper #309 64 places worse paper 0.7404 ±0.0040 best today: SpeGCL: Self-supervised Graph Spectrum Contrastive Learning without Positive Samples (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.7404 ±0.0040 Δ today 0.0098
251	MVGRL GNN baseline actual #251 vs paper #309 58 places worse	0.7500 ±0.0290 best reported	actual #251 vs paper #309 58 places worse paper 0.7402 ±0.0030 best today: Balancing Graph Embedding Smoothness in Self-Supervised Learning via Information-Theoretic Decomposition (WWW 2025) Difference within reported uncertainty; not counted as underperformance. paper 0.7402 ±0.0030 Δ today 0.0098
265	GraphCL GNN baseline actual #265 vs paper #296 31 places worse	0.7489 ±0.0065 best reported	actual #265 vs paper #296 31 places worse paper 0.7439 ±0.0040 best at pub: Graph Masked Autoencoders with Transformers (2022) Difference within reported uncertainty; not counted as underperformance. paper 0.7439 ±0.0040 Δ at pub 0.0050
280	CI+AD-GCL Hyb baseline	0.7466 ±0.0060	This paper
283	JOAO GNN baseline actual #283 vs paper #288 5 places worse	0.7460 ±0.0040 best reported	actual #283 vs paper #288 5 places worse paper 0.7455 ±0.0040 best at pub: GPS: graph contrastive learning via multi-scale augmented views from adversarial pooling (2024) best today: Balancing Graph Embedding Smoothness in Self-Supervised Learning via Information-Theoretic Decomposition (WWW 2025) Difference within reported uncertainty; not counted as underperformance. paper 0.7455 ±0.0040 Δ at pub 0.0005 Δ today 0.0005
294	InfoGraph GNN baseline	0.7444 ±0.0030	This paper
444	CI+AutoGCL Hyb baseline	0.7081 ±0.0050	This paper

COLLAB

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	WL-MLP GT global	0.9792	1-WL Expressiveness Is (Almost) All You Need
2	MSH-GNN GNN global	0.9640 ±0.0070	Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing
3	CoCN code ↗ GNN global	0.8722 ±0.0013	Scalable Graph Compressed Convolutions code ↗
23	GCN code ↗ GNN global	0.8260 ±0.0220	Scalable Expressiveness through Preprocessed Graph Perturbations code ↗
146	InfoGraph GNN baseline actual #146 vs paper #295 149 places worse	0.7940 best reported	actual #146 vs paper #295 149 places worse paper 0.7065 ±0.0110 best at pub: CogDL: A Comprehensive Library for Graph Deep Learning (WWW 2021) paper 0.7065 ±0.0110 Δ at pub 0.0875
151	MLP Trad global	0.7920 ±0.0240	Simplifying Graph Convolutional Networks with Redundancy-Free Neighbors
223	GCL-SPAN GNN baseline actual #223 vs paper #289 66 places worse	0.7637 ±0.0073 best reported	actual #223 vs paper #289 66 places worse paper 0.7140 ±0.0050 best at pub: Spectral Augmentation for Self-Supervised Learning on Graphs (ICLR 2022) paper 0.7140 ±0.0050 Δ at pub 0.0497
239	GCS GNN baseline actual #239 vs paper #262 23 places worse	0.7556 ±0.0041 best reported	actual #239 vs paper #262 23 places worse paper 0.7400 ±0.0040 best today: SpeGCL: Self-supervised Graph Spectrum Contrastive Learning without Positive Samples (2024) paper 0.7400 ±0.0040 Δ today 0.0156
264	CI+AD-GCL Hyb baseline	0.7394 ±0.0030	This paper
283	CI+AutoGCL Hyb baseline	0.7224 ±0.0090	This paper

RDT-B

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	OEPG Hyb global	0.9630 ±0.0090	Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning
2	MSH-GNN GNN global	0.9580 ±0.0090	Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing
3	GNN-AK+ code ↗ GNN global	0.9480 ±0.0080	From Stars to Subgraphs: Uplifting Any GNN with Local Structure Awareness code ↗
13	GIN GNN global	0.9240 ±0.0250	Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing
23	GraphCL GNN baseline actual #23 vs paper #43 20 places worse	0.9169 ±0.0070 best reported	actual #23 vs paper #43 20 places worse paper 0.8953 ±0.0080 best at pub: Graph Contrastive Learning with Cohesive Subgraph Awareness (WWW 2024) paper 0.8953 ±0.0080 Δ at pub 0.0216
30	CI-GCL GNN ★ proposed	0.9080 ±0.0050	This paper
36	SEGA GNN baseline	0.9021 ±0.0070	This paper
47	AutoGCL GNN baseline actual #47 vs paper #56 9 places worse	0.8931 ±0.0148 best reported	actual #47 vs paper #56 9 places worse paper 0.8660 ±0.0150 best at pub: Graph Contrastive Learning with Cohesive Subgraph Awareness (WWW 2024) paper 0.8660 ±0.0150 Δ at pub 0.0271
54	CI+AD-GCL Hyb baseline	0.8780 ±0.0120	This paper
56	CI+AutoGCL Hyb baseline	0.8750 ±0.0140	This paper
57	GCS GNN baseline	0.8650 ±0.0030	This paper
58	GCL-SPAN GNN baseline	0.8650 ±0.0010	This paper

MoleculeNet

Tox21

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	HIMP Hyb global	0.8740 ±0.0050	Graph convolutions that can finally model local structure
2	GINE w/ VN GNN global	0.8720 ±0.0030	Graph convolutions that can finally model local structure
3	NaiveGINe+ K=3 w/ VN GNN global	0.8700 ±0.0040	Graph convolutions that can finally model local structure
14	GCN GNN global	0.8400 ±0.0040	Graph convolutions that can finally model local structure
31	JOAO GNN baseline actual #31 vs paper #293 262 places worse	0.8230 ±0.0220 best reported	actual #31 vs paper #293 262 places worse paper 0.7183 ±0.0092 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.7183 ±0.0092 Δ at pub 0.1047
47	GraphCL GNN baseline actual #47 vs paper #287 240 places worse	0.8050 ±0.0170 best reported	actual #47 vs paper #287 240 places worse paper 0.7240 ±0.0101 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.7240 ±0.0101 Δ at pub 0.0810
81	InfoGraph GNN baseline actual #81 vs paper #305 224 places worse	0.7930 best reported	actual #81 vs paper #305 224 places worse paper 0.6974 ±0.0057 best at pub: GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction (AAAI 2021) paper 0.6974 ±0.0057 Δ at pub 0.0956
165	SEGA ext data GNN baseline	0.7670 ±0.0040	This paper
204	AutoGCL ext data GNN baseline	0.7550 ±0.0030	This paper
240	AD-GCL GNN baseline actual #240 vs paper #295 55 places worse	0.7470 ±0.0110 best reported	actual #240 vs paper #295 55 places worse paper 0.7142 ±0.0073 best at pub: Decoupling Weighing and Selecting for Integrating Multiple Graph Pre-training Tasks (ICLR 2024) paper 0.7142 ±0.0073 Δ at pub 0.0328
250	GCS ext data GNN baseline	0.7440 ±0.0040	This paper
264	No Pre-train GNN baseline	0.7400 ±0.0080	This paper
272	CI-GCL GNN ★ proposed	0.7359 ±0.0066	This paper
286	GCL-SPAN GNN baseline	0.7283 ±0.0062	This paper
301	MVGRL GNN baseline	0.7048 ±0.0083	This paper

BACE

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	ProtoMol Hyb global	0.9140 ±0.0030	ProtoMol: Enhancing Molecular Property Prediction via Prototype-Guided Multimodal Learning
2	MMSG GNN global	0.9080 ±0.0050	ProtoMol: Enhancing Molecular Property Prediction via Prototype-Guided Multimodal Learning
3	MemGNN code ↗ Hyb global	0.9070	Memory-Based Graph Networks code ↗
43	GIN code ↗ GNN global	0.8646 ±0.0081	Pre-training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information … code ↗
87	CI-GCL ext data GNN ★ proposed	0.8440 ±0.0090	This paper
100	JOAO ext data GNN baseline actual #100 vs paper #226 126 places worse	0.8340 ±0.0330 best reported	actual #100 vs paper #226 126 places worse paper 0.7730 ±0.0050 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.7730 ±0.0050 Δ at pub 0.0610
124	GCS ext data GNN baseline	0.8230 ±0.0030	This paper
177	GCL-SPAN ext data GNN baseline	0.7990 ±0.0070	This paper
202	AD-GCL ext data GNN baseline actual #202 vs paper #202	0.7900 ±0.0080	actual #202 vs paper #202 paper 0.7900 ±0.0080 prev best: Decoupling Weighing and Selecting for Integrating Multiple Graph Pre-training Tasks (ICLR 2024) paper 0.7900 ±0.0080 Δ at pub 0.0270
228	MVGRL ext data GNN baseline	0.7720 ±0.0100	This paper
229	SEGA ext data GNN baseline	0.7710 ±0.0050	This paper
235	AutoGCL ext data GNN baseline	0.7670 ±0.0110	This paper
281	No Pre-train GNN baseline	0.7010 ±0.0540	This paper

BBBP

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	MAGPrompt+ Hyb global	0.9432 ±0.0238	MAGPrompt: Message-Adaptive Graph Prompt Tuning for Graph Neural Networks
2	CamS-LLaMA LLM 100.0M global	0.9420	Multi-scale Graph Autoregressive Modeling: Molecular Property Prediction via Next Token …
3	SE(3)-I MPPH Hyb global	0.9400 ±0.0210	SE(3)-Invariant Multiparameter Persistent Homology for Chiral-Sensitive Molecular Property …
34	GIN code ↗ GNN global	0.8875 ±0.0049	Pre-training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information … code ↗
70	CI-GCL ext data GNN ★ proposed	0.7440 ±0.0190	This paper
91	GCS ext data GNN baseline	0.7250 ±0.0050	This paper
102	AutoGCL ext data GNN baseline	0.7200 ±0.0060	This paper
107	SEGA ext data GNN baseline	0.7190 ±0.0110	This paper
144	AD-GCL ext data GNN baseline actual #144 vs paper #223 79 places worse	0.7010 ±0.0120 best reported	actual #144 vs paper #223 79 places worse paper 0.6740 ±0.0100 best at pub: Decoupling Weighing and Selecting for Integrating Multiple Graph Pre-training Tasks (ICLR 2024) paper 0.6740 ±0.0100 Δ at pub 0.0270
152	GCL-SPAN ext data GNN baseline	0.7000 ±0.0070	This paper
180	MVGRL ext data GNN baseline	0.6900 ±0.0050	This paper

ClinTox

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	SE(3)-I MPPH Hyb global	0.9930 ±0.0040	SE(3)-Invariant Multiparameter Persistent Homology for Chiral-Sensitive Molecular Property …
2	KA-GCN code ↗ GNN global	0.9920 ±0.0050	A Systematic Survey and Benchmark of Deep Learning for Molecular Property Prediction in the … code ↗
3	KA-GAT code ↗ GNN global	0.9910 ±0.0050	A Systematic Survey and Benchmark of Deep Learning for Molecular Property Prediction in the … code ↗
19	GCN GNN global	0.9130 ±0.0170	OOD-GNN: Out-of-Distribution Generalized Graph Neural Network
58	SEGA ext data GNN baseline	0.8500 ±0.0090	This paper
79	JOAO GNN baseline actual #79 vs paper #104 25 places worse	0.8132 ±0.0249 best reported	actual #79 vs paper #104 25 places worse paper 0.7821 ±0.0412 best at pub: Graph Self-Contrast Representation Learning (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.7821 ±0.0412 Δ at pub 0.0311
82	AD-GCL GNN baseline actual #82 vs paper #82	0.8077 ±0.0392	actual #82 vs paper #82 paper 0.8077 ±0.0392 prev best: Decoupling Weighing and Selecting for Integrating Multiple Graph Pre-training Tasks (ICLR 2024) Improvement within reported uncertainty; not highlighted. paper 0.8077 ±0.0392 Δ at pub 0.0267
86	GCL-SPAN GNN baseline	0.8028 ±0.0242	This paper
90	AutoGCL ext data GNN baseline	0.7990 ±0.0330	This paper
100	CI-GCL GNN ★ proposed	0.7890 ±0.0259	This paper
97	GraphCL GNN baseline actual #97 vs paper #130 33 places worse	0.7890 ±0.0420 best reported	actual #97 vs paper #130 33 places worse paper 0.7492 ±0.0442 best at pub: 16 Pre-training Molecular Graph Representation with 3D Geometry (ICLR 2021) best today: Improving Molecular Pretraining with Complementary Featurizations (2022) Difference within reported uncertainty; not counted as underperformance. paper 0.7492 ±0.0442 Δ at pub 0.0398 Δ today 0.0398
106	InfoGraph GNN baseline actual #106 vs paper #204 98 places worse	0.7810 best reported	actual #106 vs paper #204 98 places worse paper 0.6450 ±0.0532 best at pub: GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction (AAAI 2021) paper 0.6450 ±0.0532 Δ at pub 0.1360
141	MVGRL GNN baseline	0.7384 ±0.0425	This paper
195	GCS ext data GNN baseline	0.6670 ±0.0190	This paper
232	No Pre-train GNN baseline	0.5800 ±0.0440	This paper

SIDER

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	KA-GAT code ↗ GNN global	0.8470 ±0.0020	A Systematic Survey and Benchmark of Deep Learning for Molecular Property Prediction in the … code ↗
2	KA-GCN code ↗ GNN global	0.8420 ±0.0010	A Systematic Survey and Benchmark of Deep Learning for Molecular Property Prediction in the … code ↗
3	GraphKAN code ↗ GNN global	0.8370 ±0.0010	A Systematic Survey and Benchmark of Deep Learning for Molecular Property Prediction in the … code ↗
51	GCL-SPAN GNN baseline	0.6487 ±0.0088	This paper
55	CI-GCL GNN ★ proposed	0.6470 ±0.0030	This paper
66	GIN code ↗ GNN global	0.6403 ±0.0104	Pre-training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information … code ↗
68	JOAO GNN baseline actual #68 vs paper #90 22 places worse	0.6400 ±0.0100 best reported	actual #68 vs paper #90 22 places worse paper 0.6273 ±0.0092 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) Difference within reported uncertainty; not counted as underperformance. paper 0.6273 ±0.0092 Δ at pub 0.0127
74	SEGA ext data GNN baseline	0.6370 ±0.0030	This paper
98	AutoGCL ext data GNN baseline	0.6250 ±0.0060	This paper
103	ContextPred ext data GNN baseline actual #103 vs paper #144 41 places worse	0.6220 ±0.0059 best reported	actual #103 vs paper #144 41 places worse paper 0.6090 ±0.0060 best at pub: A Group Symmetric Stochastic Differential Equation Model for Molecule Multi-modal Pretraining (ICML 2023) best today: A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery (2025) paper 0.6090 ±0.0060 Δ at pub 0.0130 Δ today 0.0130
110	MVGRL GNN baseline	0.6194 ±0.0094	This paper
111	GCS ext data GNN baseline	0.6190 ±0.0040	This paper
114	GraphCL GNN baseline actual #114 vs paper #116 2 places worse	0.6176 ±0.0111	actual #114 vs paper #116 2 places worse paper 0.6176 ±0.0111 prev best: Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning (ICML 2022) Improvement within reported uncertainty; not highlighted. paper 0.6176 ±0.0111 Δ at pub 0.0096
120	Infomax ext data GNN baseline actual #120 vs paper #206 86 places worse	0.6160 ±0.0240 best reported	actual #120 vs paper #206 86 places worse paper 0.5840 ±0.0080 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) Difference within reported uncertainty; not counted as underperformance. paper 0.5840 ±0.0080 Δ at pub 0.0320
133	AttrMasking ext data GNN baseline actual #133 vs paper #141 8 places worse	0.6121 ±0.0065 best reported	actual #133 vs paper #141 8 places worse paper 0.6100 ±0.0070 best at pub: GraphMAE: Self-Supervised Masked Graph Autoencoders (KDD 2022) best today: Pre-training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck (AAAI 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.6100 ±0.0070 Δ today 0.0021
146	AD-GCL GNN baseline actual #146 vs paper #146	0.6080 ±0.0090	actual #146 vs paper #146 paper 0.6080 ±0.0090 prev best: Decoupling Weighing and Selecting for Integrating Multiple Graph Pre-training Tasks (ICLR 2024) paper 0.6080 ±0.0090 Δ at pub 0.0140
162	InfoGraph GNN baseline actual #162 vs paper #162	0.6054 ±0.0090	actual #162 vs paper #162 paper 0.6054 ±0.0090 prev best: A Group Symmetric Stochastic Differential Equation Model for Molecule Multi-modal Pretraining (ICML 2023) Improvement within reported uncertainty; not highlighted. paper 0.6054 ±0.0090 Δ at pub 0.0064
170	EdgePred ext data GNN baseline	0.6040 ±0.0070	This paper
222	No Pre-train GNN baseline	0.5730 ±0.0160	This paper

ToxCast

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	GeomGCL GNN global	0.7630	GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction
2	LiGhT DeepGPT ext data Hyb 0.4M global	0.7570 ±0.0110	Deep Prompt Tuning for Graph Transformers
3	CoMPT GT global	0.7550	GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction
14	GCN GNN global	0.7350 ±0.0020	Graph convolutions that can finally model local structure
31	JOAO ext data GNN baseline actual #31 vs paper #165 134 places worse	0.7110 ±0.0140 best reported	actual #31 vs paper #165 134 places worse paper 0.6290 ±0.0050 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.6290 ±0.0050 Δ at pub 0.0820
100	CI-GCL ext data GNN ★ proposed	0.6540 ±0.0150	This paper
104	SEGA ext data GNN baseline	0.6520 ±0.0090	This paper
118	GCS ext data GNN baseline	0.6440 ±0.0020	This paper
127	GCL-SPAN ext data GNN baseline	0.6420 ±0.0040	This paper
138	AD-GCL ext data GNN baseline actual #138 vs paper #138	0.6350 ±0.0070	actual #138 vs paper #138 paper 0.6350 ±0.0070 prev best: Let Invariant Rationale Discovery Inspire Graph Contrastive Learning (ICML 2022) Improvement within reported uncertainty; not highlighted. paper 0.6350 ±0.0070 Δ at pub 0.0040
142	No Pre-train GNN baseline	0.6340 ±0.0060	This paper
143	AutoGCL ext data GNN baseline	0.6340 ±0.0040	This paper
176	MVGRL ext data GNN baseline	0.6260 ±0.0050	This paper

FreeSolv

Global leaderboard →

#	Model	RMSE ↓	Paper
1	SE(3)-I MPPH Hyb global	0.3540 ±0.0530	SE(3)-Invariant Multiparameter Persistent Homology for Chiral-Sensitive Molecular Property …
2	ECFP-4 + RF RF global	0.5600 ±0.0660	SE(3)-Invariant Multiparameter Persistent Homology for Chiral-Sensitive Molecular Property …
3	GPSE + GPS code ↗ ext data Hyb global	0.6820	Graph Positional and Structural Encoder code ↗
56	InfoGraph GNN baseline actual #56 vs paper #128 72 places worse	2.1040 best reported	actual #56 vs paper #128 72 places worse paper 10.0000 ±4.8200 best at pub: GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction (AAAI 2021) paper 10.0000 ±4.8200 Δ at pub 7.8960
69	GIN code ↗ GNN global	2.3460 ±0.1220	ChemRL-GEM: Geometry Enhanced Molecular Representation Learning for Property Prediction code ↗
91	CI-GCL GNN ★ proposed	2.8730 ±0.3200	This paper
101	GraphCL GNN baseline actual #101 vs paper #125 24 places worse	3.1660 ±0.0270 best reported	actual #101 vs paper #125 24 places worse paper 7.6790 ±2.7500 best at pub: 3D Infomax improves GNNs for Molecular Property Prediction (ICML 2021) best today: Hierarchical Molecular Representation Learning via Fragment-Based Self-Supervised Embedding Prediction (2026) paper 7.6790 ±2.7500 Δ at pub 3.9350 Δ today 4.5130
104	JOAO GNN baseline actual #104 vs paper #123 19 places worse	3.2820 ±0.0020 best reported	actual #104 vs paper #123 19 places worse paper 5.1310 ±0.7200 best at pub: Motif-aware Attribute Masking for Molecular Graph Pre-training (2023) best today: Hierarchical Molecular Representation Learning via Fragment-Based Self-Supervised Embedding Prediction (2026) paper 5.1310 ±0.7200 Δ at pub 1.6640 Δ today 1.8490
121	GCL-SPAN GNN baseline	4.5310 ±0.4600	This paper
124	AD-GCL GNN baseline	5.1500 ±0.6200	This paper
128	MVGRL GNN baseline	9.0240 ±1.9800	This paper

SOTA performance (not significantly different from the best) Proposed method (this paper) Baseline / comparison (this paper) Global context (other papers) Best at pub — reported result worse than the best known when published (best reported in green) Best at pub — reported value improved over the best known when published Best today — OK at publication; today's leaderboard best is elsewhere