Improving Molecular Pretraining with Complementary Featurizations

Yanqiao Zhu, Dingshuo Chen, Yuanqi Du, Yingze Wang, Q. Liu, Shu Wu

Abstract

-0.5em Molecular pretraining, which learns molecular representations over massive unlabeled data, has become a prominent paradigm to solve a variety of tasks in computational chemistry and drug discovery. Recently, prosperous progress has been made in molecular pretraining with different molecular featurizations, including 1D SMILES strings, 2D graphs, and 3D geometries. However, the role of molecular featurizations with their corresponding neural architectures in molecular pretraining remains largely unexamined. In this paper, through two case studies---chirality classification and aromatic ring counting---we first demonstrate that different featurization techniques convey chemical information differently. In light of this observation, we propose a simple and effective lecular pretraining framework with mplementary featurizations (MOCO). MOCO comprehensively leverages multiple featurizations that complement each other and outperforms existing state-of-the-art models that solely relies on one or two featurizations on a wide range of molecular property prediction tasks.

Benchmark Results

Tested on 8 datasets 29 baseline values reported too low

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

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 #247 216 places worse	0.8230 ±0.0220 best reported	actual #31 vs paper #247 216 places worse paper 0.7440 ±0.0070 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.7440 ±0.0070 Δ at pub 0.0790
46	ContextPred GNN baseline actual #46 vs paper #279 233 places worse	0.8060 ±0.0120 best reported	actual #46 vs paper #279 233 places worse paper 0.7330 ±0.0050 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.7330 ±0.0050 Δ at pub 0.0730
47	GraphCL GNN baseline actual #47 vs paper #231 184 places worse	0.8050 ±0.0170 best reported	actual #47 vs paper #231 184 places worse paper 0.7500 ±0.0030 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.7500 ±0.0030 Δ at pub 0.0550
61	GraphMVP GNN baseline actual #61 vs paper #246 185 places worse	0.7990 ±0.0180 best reported	actual #61 vs paper #246 185 places worse paper 0.7450 ±0.0040 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.7450 ±0.0040 Δ at pub 0.0540
71	GraphLoG GNN baseline actual #71 vs paper #282 211 places worse	0.7960 ±0.0250 best reported	actual #71 vs paper #282 211 places worse paper 0.7300 ±0.0030 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.7300 ±0.0030 Δ at pub 0.0660
81	InfoGraph GNN baseline actual #81 vs paper #282 201 places worse	0.7930 best reported	actual #81 vs paper #282 201 places worse paper 0.7300 ±0.0070 best at pub: GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction (AAAI 2021) paper 0.7300 ±0.0070 Δ at pub 0.0630
135	GraphSAGE GNN baseline actual #135 vs paper #246 111 places worse	0.7744 ±0.0038 best reported	actual #135 vs paper #246 111 places worse paper 0.7450 ±0.0040 best today: Pre-training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck (AAAI 2024) paper 0.7450 ±0.0040 Δ today 0.0294
164	MOCO ext data Hyb ★ proposed	0.7670 ±0.0040	This paper
180	GPT-GNN LLM baseline actual #180 vs paper #211 31 places worse	0.7610 ±0.0040 best reported	actual #180 vs paper #211 31 places worse paper 0.7530 ±0.0050 best at pub: Motif-based Graph Self-Supervised Learning for Molecular Property Prediction (NeurIPS 2021) paper 0.7530 ±0.0050 Δ at pub 0.0080
185	--- Trad baseline	0.7590 ±0.0030	This paper
202	GraphMAE GNN baseline actual #202 vs paper #231 29 places worse	0.7550 ±0.0060 best reported	actual #202 vs paper #231 29 places worse paper 0.7500 ±0.0040 best at pub: GraphMAE: Self-Supervised Masked Graph Autoencoders (KDD 2022) best today: A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.7500 ±0.0040 Δ at pub 0.0050 Δ today 0.0050
222	GROVER-C GNN baseline	0.7520 ±0.0030	This paper
229	AttrMask GNN baseline actual #229 vs paper #259 30 places worse	0.7510 ±0.0090 best reported	actual #229 vs paper #259 30 places worse paper 0.7420 ±0.0080 best at pub: 16 Pre-training Molecular Graph Representation with 3D Geometry (ICLR 2021) best today: Motif-aware Attribute Masking for Molecular Graph Pre-training (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.7420 ±0.0080 Δ today 0.0090
281	GROVER-M GNN baseline	0.7320 ±0.0080	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 ↗
82	GraphSAGE GNN baseline actual #82 vs paper #295 213 places worse	0.8451 best reported	actual #82 vs paper #295 213 places worse paper 0.6460 ±0.0470 best today: Application of Graph Neural Networks and graph descriptors for graph classification (2022) paper 0.6460 ±0.0470 Δ today 0.1991
98	GraphLoG ext data GNN baseline actual #98 vs paper #206 108 places worse	0.8350 ±0.0120 best reported	actual #98 vs paper #206 108 places worse paper 0.7880 ±0.0070 best at pub: Let Invariant Rationale Discovery Inspire Graph Contrastive Learning (ICML 2022) best today: UGMAE: A Unified Framework for Graph Masked Autoencoders (2024) paper 0.7880 ±0.0070 Δ at pub 0.0380 Δ today 0.0470
100	JOAO ext data GNN baseline actual #100 vs paper #261 161 places worse	0.8340 ±0.0330 best reported	actual #100 vs paper #261 161 places worse paper 0.7290 ±0.0200 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.7290 ±0.0200 Δ at pub 0.1050
105	GraphMAE ext data GNN baseline actual #105 vs paper #141 36 places worse	0.8310 ±0.0090 best reported	actual #105 vs paper #141 36 places worse paper 0.8140 ±0.0140 best at pub: GraphMAE: Self-Supervised Masked Graph Autoencoders (KDD 2022) best today: A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.8140 ±0.0140 Δ at pub 0.0170 Δ today 0.0170
117	MOCO ext data Hyb ★ proposed	0.8260 ±0.0030	This paper
149	GraphMVP ext data GNN baseline actual #149 vs paper #232 83 places worse	0.8120 ±0.0090 best reported	actual #149 vs paper #232 83 places worse paper 0.7680 ±0.0110 best at pub: Unified 2D and 3D Pre-Training of Molecular Representations (KDD 2022) best today: Invariant-Based Diagnostics for Graph Benchmarks (2026) paper 0.7680 ±0.0110 Δ at pub 0.0440 Δ today 0.0440
182	AttrMask ext data GNN baseline actual #182 vs paper #226 44 places worse	0.7970 ±0.0033 best reported	actual #182 vs paper #226 44 places worse paper 0.7720 ±0.0140 best at pub: 16 Pre-training Molecular Graph Representation with 3D Geometry (ICLR 2021) best today: A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery (2025) paper 0.7720 ±0.0140 Δ today 0.0250
186	ContextPred ext data GNN baseline actual #186 vs paper #207 21 places worse	0.7960 ±0.0120 best reported	actual #186 vs paper #207 21 places worse paper 0.7860 ±0.0140 best at pub: 16 Pre-training Molecular Graph Representation with 3D Geometry (ICLR 2021) best today: Dual-perspective Cross Contrastive Learning in Graph Transformers (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.7860 ±0.0140 Δ today 0.0100
197	GROVER-C ext data GNN baseline	0.7920 ±0.0030	This paper
204	InfoGraph ext data GNN baseline actual #204 vs paper #255 51 places worse	0.7890 ±0.0110 best reported	actual #204 vs paper #255 51 places worse paper 0.7390 ±0.0250 best at pub: Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning (ICML 2022) paper 0.7390 ±0.0250 Δ at pub 0.0500
208	GPT-GNN ext data LLM baseline actual #208 vs paper #225 17 places worse	0.7860 ±0.0290 best reported	actual #208 vs paper #225 17 places worse paper 0.7760 ±0.0050 best at pub: Motif-based Graph Self-Supervised Learning for Molecular Property Prediction (NeurIPS 2021) best today: Decoupling Weighing and Selecting for Integrating Multiple Graph Pre-training Tasks (ICLR 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.7760 ±0.0050 Δ at pub 0.0090 Δ today 0.0100
213	GraphCL ext data GNN baseline actual #213 vs paper #286 73 places worse	0.7820 ±0.0120 best reported	actual #213 vs paper #286 73 places worse paper 0.6870 ±0.0780 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 0.6870 ±0.0780 Δ at pub 0.0950
260	GROVER-M ext data GNN baseline	0.7340 ±0.0400	This paper
271	--- Trad baseline	0.7150 ±0.0270	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 ↗
36	GraphCL GNN baseline actual #36 vs paper #219 183 places worse	0.8870 ±0.0190 best reported	actual #36 vs paper #219 183 places worse paper 0.6750 ±0.0330 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.6750 ±0.0330 Δ at pub 0.2120
39	JOAO GNN baseline actual #39 vs paper #253 214 places worse	0.8790 ±0.0200 best reported	actual #39 vs paper #253 214 places worse paper 0.6600 ±0.0060 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.6600 ±0.0060 Δ at pub 0.2190
40	ContextPred GNN baseline actual #40 vs paper #118 78 places worse	0.8770 ±0.0260 best reported	actual #40 vs paper #118 78 places worse paper 0.7120 ±0.0090 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.7120 ±0.0090 Δ at pub 0.1650
49	GraphMVP GNN baseline actual #49 vs paper #190 141 places worse	0.8600 ±0.0340 best reported	actual #49 vs paper #190 141 places worse paper 0.6850 ±0.0020 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.6850 ±0.0020 Δ at pub 0.1750
54	GraphLoG GNN baseline actual #54 vs paper #205 151 places worse	0.8460 ±0.0080 best reported	actual #54 vs paper #205 151 places worse paper 0.6780 ±0.0170 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.6780 ±0.0170 Δ at pub 0.1680
57	GraphSAGE GNN baseline actual #57 vs paper #278 221 places worse	0.8304 ±0.0038 best reported	actual #57 vs paper #278 221 places worse paper 0.6450 ±0.0310 best today: Pre-training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck (AAAI 2024) paper 0.6450 ±0.0310 Δ today 0.1854
101	GraphMAE GNN baseline actual #101 vs paper #123 22 places worse	0.7200 ±0.0060 best reported	actual #101 vs paper #123 22 places worse paper 0.7090 ±0.0090 best at pub: GraphMAE: Self-Supervised Masked Graph Autoencoders (KDD 2022) best today: A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.7090 ±0.0090 Δ at pub 0.0110 Δ today 0.0110
112	MOCO ext data Hyb ★ proposed	0.7160 ±0.0100	This paper
122	--- Trad baseline	0.7100 ±0.0050	This paper
138	GROVER-C GNN baseline	0.7030 ±0.0160	This paper
141	AttrMask GNN baseline	0.7020 ±0.0050	This paper
173	InfoGraph GNN baseline	0.6920 ±0.0080	This paper
220	GPT-GNN LLM baseline actual #220 vs paper #278 58 places worse	0.6750 ±0.0130 best reported	actual #220 vs paper #278 58 places worse paper 0.6450 ±0.0110 best at pub: Motif-based Graph Self-Supervised Learning for Molecular Property Prediction (NeurIPS 2021) paper 0.6450 ±0.0110 Δ at pub 0.0300
246	GROVER-M GNN baseline	0.6640 ±0.0340	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
40	AttrMask GNN baseline actual #40 vs paper #183 143 places worse	0.8770 ±0.0119 best reported	actual #40 vs paper #183 143 places worse paper 0.6860 ±0.0960 best at pub: 16 Pre-training Molecular Graph Representation with 3D Geometry (ICLR 2021) best today: A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery (2025) paper 0.6860 ±0.0960 Δ today 0.1910
49	GraphMVP GNN baseline actual #49 vs paper #96 47 places worse	0.8650 ±0.0170 best reported	actual #49 vs paper #96 47 places worse paper 0.7900 ±0.0250 best at pub: 16 Pre-training Molecular Graph Representation with 3D Geometry (ICLR 2021) best today: SubGDiff: A Subgraph Diffusion Model to Improve Molecular Representation Learning (NeurIPS 2024) paper 0.7900 ±0.0250 Δ today 0.0750
71	GraphMAE GNN baseline actual #71 vs paper #77 6 places worse	0.8230 ±0.0120 best reported	actual #71 vs paper #77 6 places worse paper 0.8150 ±0.0280 best at pub: GraphMAE: Self-Supervised Masked Graph Autoencoders (KDD 2022) best today: A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.8150 ±0.0280 Δ at pub 0.0080 Δ today 0.0080
77	MOCO ext data Hyb ★ proposed	0.8160 ±0.0370	This paper
79	JOAO GNN baseline actual #79 vs paper #195 116 places worse	0.8132 ±0.0249 best reported	actual #79 vs paper #195 116 places worse paper 0.6630 ±0.0390 best at pub: GraphMAE: Self-Supervised Masked Graph Autoencoders (KDD 2022) best today: Graph Self-Contrast Representation Learning (2023) paper 0.6630 ±0.0390 Δ at pub 0.1500 Δ today 0.1502
97	GraphCL GNN baseline	0.7890 ±0.0420	This paper
103	GraphLoG GNN baseline actual #103 vs paper #216 113 places worse	0.7872 ±0.0258 best reported	actual #103 vs paper #216 113 places worse paper 0.6200 ±0.0180 best at pub: Let Invariant Rationale Discovery Inspire Graph Contrastive Learning (ICML 2022) best today: Graph Self-Contrast Representation Learning (2023) paper 0.6200 ±0.0180 Δ at pub 0.1672 Δ today 0.1672
106	InfoGraph GNN baseline actual #106 vs paper #128 22 places worse	0.7810 best reported	actual #106 vs paper #128 22 places worse paper 0.7510 ±0.0500 best at pub: GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction (AAAI 2021) Difference within reported uncertainty; not counted as underperformance. paper 0.7510 ±0.0500 Δ at pub 0.0300
107	GROVER-M GNN baseline	0.7780 ±0.0200	This paper
115	ContextPred GNN baseline actual #115 vs paper #142 27 places worse	0.7720 ±0.0088 best reported	actual #115 vs paper #142 27 places worse paper 0.7370 ±0.0400 best at pub: 16 Pre-training Molecular Graph Representation with 3D Geometry (ICLR 2021) best today: A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.7370 ±0.0400 Δ today 0.0350
131	GPT-GNN LLM baseline actual #131 vs paper #232 101 places worse	0.7490 ±0.0270 best reported	actual #131 vs paper #232 101 places worse paper 0.5780 ±0.0310 best at pub: Motif-based Graph Self-Supervised Learning for Molecular Property Prediction (NeurIPS 2021) paper 0.5780 ±0.0310 Δ at pub 0.1710
165	--- Trad baseline	0.7150 ±0.0530	This paper
223	GROVER-C GNN baseline	0.5990 ±0.0820	This paper
238	GraphSAGE GNN baseline	0.5580 ±0.0620	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 ↗
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 #150 82 places worse	0.6400 ±0.0100 best reported	actual #68 vs paper #150 82 places worse paper 0.6070 ±0.0100 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.6070 ±0.0100 Δ at pub 0.0330
70	GraphMVP GNN baseline actual #70 vs paper #99 29 places worse	0.6390 ±0.0120 best reported	actual #70 vs paper #99 29 places worse paper 0.6230 ±0.0160 best at pub: Unified 2D and 3D Pre-Training of Molecular Representations (KDD 2022) best today: A Systematic Survey and Benchmark of Deep Learning for Molecular Property Prediction in the Foundation Model Era (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.6230 ±0.0160 Δ at pub 0.0160 Δ today 0.0160
103	ContextPred GNN baseline actual #103 vs paper #196 93 places worse	0.6220 ±0.0059 best reported	actual #103 vs paper #196 93 places worse paper 0.5930 ±0.0140 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) best today: A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery (2025) paper 0.5930 ±0.0140 Δ at pub 0.0160 Δ today 0.0290
114	GraphCL GNN baseline actual #114 vs paper #178 64 places worse	0.6183 ±0.0060 best reported	actual #114 vs paper #178 64 places worse paper 0.6010 ±0.0130 best at pub: Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning (ICML 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 0.6010 ±0.0130 Δ at pub 0.0070 Δ today 0.0173
123	GraphLoG GNN baseline actual #123 vs paper #220 97 places worse	0.6150 ±0.0130 best reported	actual #123 vs paper #220 97 places worse paper 0.5740 ±0.0230 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.5740 ±0.0230 Δ at pub 0.0410
136	MOCO ext data Hyb ★ proposed	0.6120 ±0.0060	This paper
134	AttrMask GNN baseline actual #134 vs paper #169 35 places worse	0.6120 ±0.0012 best reported	actual #134 vs paper #169 35 places worse paper 0.6040 ±0.0060 best at pub: 16 Pre-training Molecular Graph Representation with 3D Geometry (ICLR 2021) best today: A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.6040 ±0.0060 Δ today 0.0080
160	GROVER-M GNN baseline	0.6060 ±0.0110	This paper
162	InfoGraph GNN baseline actual #162 vs paper #199 37 places worse	0.6054 ±0.0090 best reported	actual #162 vs paper #199 37 places worse paper 0.5920 ±0.0020 best at pub: Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning (ICML 2022) best today: Community-Invariant Graph Contrastive Learning (ICML 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.5920 ±0.0020 Δ at pub 0.0020 Δ today 0.0134
175	GraphMAE GNN baseline actual #175 vs paper #183 8 places worse	0.6030 ±0.0110 best reported	actual #175 vs paper #183 8 places worse paper 0.5990 ±0.0050 best at pub: GraphMAE: Self-Supervised Masked Graph Autoencoders (KDD 2022) best today: A Generalist Cross-Domain Molecular Learning Framework for Structure-Based Drug Discovery (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.5990 ±0.0050 Δ at pub 0.0040 Δ today 0.0040
197	GPT-GNN LLM baseline actual #197 vs paper #219 22 places worse	0.5930 ±0.0080 best reported	actual #197 vs paper #219 22 places worse paper 0.5750 ±0.0420 best at pub: Motif-based Graph Self-Supervised Learning for Molecular Property Prediction (NeurIPS 2021) Difference within reported uncertainty; not counted as underperformance. paper 0.5750 ±0.0420 Δ at pub 0.0180
207	GROVER-C GNN baseline	0.5840 ±0.0060	This paper
218	--- Trad baseline	0.5770 ±0.0310	This paper
227	GraphSAGE GNN baseline	0.5670 ±0.0010	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
28	ContextPred GNN baseline actual #28 vs paper #167 139 places worse	0.7140 ±0.0180 best reported	actual #28 vs paper #167 139 places worse paper 0.6280 ±0.0030 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.6280 ±0.0030 Δ at pub 0.0860
31	JOAO GNN baseline actual #31 vs paper #168 137 places worse	0.7110 ±0.0140 best reported	actual #31 vs paper #168 137 places worse paper 0.6270 ±0.0060 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.6270 ±0.0060 Δ at pub 0.0840
35	InfoGraph GNN baseline actual #35 vs paper #184 149 places worse	0.7050 best reported	actual #35 vs paper #184 149 places worse paper 0.6200 ±0.0030 best at pub: GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction (AAAI 2021) paper 0.6200 ±0.0030 Δ at pub 0.0850
42	GraphCL GNN baseline actual #42 vs paper #167 125 places worse	0.6960 ±0.0230 best reported	actual #42 vs paper #167 125 places worse paper 0.6280 ±0.0020 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.6280 ±0.0020 Δ at pub 0.0680
50	GraphMVP GNN baseline actual #50 vs paper #168 118 places worse	0.6890 ±0.0100 best reported	actual #50 vs paper #168 118 places worse paper 0.6270 ±0.0010 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.6270 ±0.0010 Δ at pub 0.0620
62	GraphSAGE GNN baseline actual #62 vs paper #194 132 places worse	0.6775 ±0.0041 best reported	actual #62 vs paper #194 132 places worse paper 0.6080 ±0.0050 best today: Pre-training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck (AAAI 2024) paper 0.6080 ±0.0050 Δ today 0.0695
64	GraphLoG GNN baseline actual #64 vs paper #181 117 places worse	0.6770 ±0.0190 best reported	actual #64 vs paper #181 117 places worse paper 0.6220 ±0.0040 best at pub: KPGT: Knowledge-Guided Pre-training of Graph Transformer for Molecular Property Prediction (KDD 2022) paper 0.6220 ±0.0040 Δ at pub 0.0550
110	MOCO ext data Hyb ★ proposed	0.6490 ±0.0080	This paper
113	--- Trad baseline	0.6470 ±0.0230	This paper
128	GraphMAE GNN baseline	0.6410 ±0.0010	This paper
150	AttrMask GNN baseline actual #150 vs paper #177 27 places worse	0.6330 ±0.0060 best reported	actual #150 vs paper #177 27 places worse paper 0.6250 ±0.0040 best at pub: 16 Pre-training Molecular Graph Representation with 3D Geometry (ICLR 2021) best today: Motif-aware Attribute Masking for Molecular Graph Pre-training (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.6250 ±0.0040 Δ today 0.0080
151	GPT-GNN LLM baseline actual #151 vs paper #181 30 places worse	0.6330 ±0.0030 best reported	actual #151 vs paper #181 30 places worse paper 0.6220 ±0.0010 best at pub: Motif-based Graph Self-Supervised Learning for Molecular Property Prediction (NeurIPS 2021) best today: Decoupling Weighing and Selecting for Integrating Multiple Graph Pre-training Tasks (ICLR 2024) paper 0.6220 ±0.0010 Δ at pub 0.0090 Δ today 0.0110
174	GROVER-C GNN baseline	0.6260 ±0.0030	This paper
175	GROVER-M GNN baseline	0.6260 ±0.0050	This paper

ESOL

Global leaderboard →

#	Model	RMSE ↓	Paper
1	Graphormer-SPIS code ↗ GT global	0.4840 ±0.0050	On Structural Expressive Power of Graph Transformers code ↗
2	Graphormer-SPD code ↗ GT global	0.4920 ±0.0040	On Structural Expressive Power of Graph Transformers code ↗
3	conan-FGW global	0.5140 ±0.0190	Structure-Aware E(3)-Invariant Molecular Conformer Aggregation Networks
7	GCN code ↗ GNN global	0.5640 ±0.0150	On Structural Expressive Power of Graph Transformers code ↗
83	MOCO ext data Hyb ★ proposed	0.9840 ±0.0340	This paper
98	GraphMVP ext data GNN baseline	1.0910 ±0.0210	This paper
103	AttrMask ext data GNN baseline	1.1120 ±0.0480	This paper
105	JOAO ext data GNN baseline	1.1200 ±0.0190	This paper
114	ContextPred ext data GNN baseline	1.1960 ±0.0370	This paper
123	--- Trad baseline	1.3640 ±0.0160	This paper

Lipophilicity

Global leaderboard →

#	Model	RMSE ↓	Paper
1	MemGNN code ↗ Hyb global	0.5560 ±0.0230	Memory-Based Graph Networks code ↗
2	ProtoS-L2 Hyb 0.1M global	0.5800 ±0.0160	Optimal Transport Graph Neural Networks
3	ProtoW-L2 Hyb global	0.6040 ±0.0140	Optimal Transport Graph Neural Networks
7	GCN code ↗ GNN global	0.6550 ±0.0360	Memory-Based Graph Networks code ↗
17	ContextPred ext data GNN baseline	0.7020 ±0.0200	This paper
20	MOCO ext data Hyb ★ proposed	0.7070 ±0.0010	This paper
21	JOAO ext data GNN baseline	0.7080 ±0.0070	This paper
27	GraphMVP ext data GNN baseline	0.7180 ±0.0160	This paper
31	AttrMask ext data GNN baseline	0.7300 ±0.0040	This paper
37	--- Trad baseline	0.7360 ±0.0060	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 today — OK at publication; today's leaderboard best is elsewhere