Research map / Graph learning / Graph self-supervised learning
Evaluating Self-Supervised Learning for Molecular Graph Embeddings
Hanchen Wang, Jean Kaddour, Shengchao Liu, Jian Tang, Joan Lasenby, Qi Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Pre-train the GIN with a GSSL objective rather than random initialization | GIN molecular property prediction · Eight MPP datasets | ROC-AUC: 12.3% on BBBP; -0.3% on MUV | Improved |
| Use GraphCL pretraining instead of random initialization | GIN molecular property prediction · Eight MPP datasets, fixed embeddings | average ROC-AUC: 59.60 -> 65.04 | Improved |
| Fine-tune pretrained GNN weights along with the probe | JOAO · Eight MPP datasets | average ROC-AUC rank | Mixed |
| Choose pretraining hyperparameters for the fine-tuning evaluation rather than fixed-embedding probing | GSSL methods on MPP · Eight MPP datasets | ROC-AUC: AttrMask mask rate=0.85 and learning rate=1e-4 -> mask rate=0.50 and learning rate=5e-4 | Mixed |
| Use random initialization instead of GSSL pretraining | GIN node and pair topology probes · Generic graph properties | average rank: 1.5 node-level average rank; 2.8 pair-level average rank | Improved |
| Use random initialization instead of GSSL pretraining | GIN graph-level topology probes · Generic graph properties | diameter MSE: 177.924 | Hurt |
| Pretrain with an adjacency-prediction objective | EdgePred GIN · Generic pair topology | Link MSE: 0.078 -> 0.067 | Improved |
| Include 3D geometry during pretraining | GraphMVP GIN · Generic pair topology | Katz MSE: 0.017 | Improved |
| Use GSSL pretraining rather than random initialization | GIN substructure probes · Molecular substructure properties | benzene count MSE: 1.100 -> 0.525 | Improved |
| Prioritize substructure-aware objectives | GSSL molecular property prediction · Eight MPP datasets | ROC-AUC: Corr. 0.915; p-value 2e-4 | Improved |
| Increase embedding uniformity | GSSL molecular embeddings · MPP tasks | ROC-AUC | No clear effect |
| Use InfoGraph, which maximizes mutual information between multi-scale graph representations | GIN molecular embeddings · Eight MPP datasets | average ROC-AUC rank: 7th among 9 GSSLs | Hurt |
Closest papers by meaning
- Bridging Molecular Graphs and Large Language Models (AAAI 2025 · similarity 0.88)
- Pre-Training Graph Neural Networks on Molecules by Using Subgraph-Conditioned Graph Information Bottleneck (AAAI 2025 · similarity 0.88)
- On the Scalability of GNNs for Molecular Graphs (NeurIPS 2024 · similarity 0.88)
- MOL-Mamba: Enhancing Molecular Representation with Structural & Electronic Insights (AAAI 2025 · similarity 0.88)
- Molecular Data Programming: Towards Molecule Pseudo-labeling with Systematic Weak Supervision (CVPR 2024 · similarity 0.87)
- MAGE: Model-Level Graph Neural Networks Explanations via Motif-based Graph Generation (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
