AI Paper Atlas

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

NeurIPS 2023 · Graph learning › Graph self-supervised learning · approach: Molecular graph pretraining · 7 citations (OpenAlex)

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 changedWhereMeasuredEffect
Pre-train the GIN with a GSSL objective rather than random initializationGIN molecular property prediction · Eight MPP datasetsROC-AUC: 12.3% on BBBP; -0.3% on MUVImproved
Use GraphCL pretraining instead of random initializationGIN molecular property prediction · Eight MPP datasets, fixed embeddingsaverage ROC-AUC: 59.60 -> 65.04Improved
Fine-tune pretrained GNN weights along with the probeJOAO · Eight MPP datasetsaverage ROC-AUC rankMixed
Choose pretraining hyperparameters for the fine-tuning evaluation rather than fixed-embedding probingGSSL methods on MPP · Eight MPP datasetsROC-AUC: AttrMask mask rate=0.85 and learning rate=1e-4 -> mask rate=0.50 and learning rate=5e-4Mixed
Use random initialization instead of GSSL pretrainingGIN node and pair topology probes · Generic graph propertiesaverage rank: 1.5 node-level average rank; 2.8 pair-level average rankImproved
Use random initialization instead of GSSL pretrainingGIN graph-level topology probes · Generic graph propertiesdiameter MSE: 177.924Hurt
Pretrain with an adjacency-prediction objectiveEdgePred GIN · Generic pair topologyLink MSE: 0.078 -> 0.067Improved
Include 3D geometry during pretrainingGraphMVP GIN · Generic pair topologyKatz MSE: 0.017Improved
Use GSSL pretraining rather than random initializationGIN substructure probes · Molecular substructure propertiesbenzene count MSE: 1.100 -> 0.525Improved
Prioritize substructure-aware objectivesGSSL molecular property prediction · Eight MPP datasetsROC-AUC: Corr. 0.915; p-value 2e-4Improved
Increase embedding uniformityGSSL molecular embeddings · MPP tasksROC-AUCNo clear effect
Use InfoGraph, which maximizes mutual information between multi-scale graph representationsGIN molecular embeddings · Eight MPP datasetsaverage ROC-AUC rank: 7th among 9 GSSLsHurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy