Convolutional Networks on Graphs for Learning Molecular Fingerprints

We introduce a convolutional neural network that operates directly on graphs. These networks allow end-to-end learning of prediction pipelines whose inputs are graphs of arbitrary size and shape. The architecture we present generalizes standard molecular feature extraction methods based on circular fingerprints. We show that these data-driven features are more interpretable, and have better predictive performance on a variety of tasks.

PDF Abstract NeurIPS 2015 PDF NeurIPS 2015 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Node Classification CiteSeer (0.5%) GCN-FP Accuracy 43.9% # 12
Node Classification CiteSeer (1%) GCN-FP Accuracy 54.3% # 13
Node Classification CiteSeer with Public Split: fixed 20 nodes per class GCN-FP Accuracy 61.5% # 34
Node Classification Cora (0.5%) GCN-FP Accuracy 50.5% # 12
Node Classification Cora (1%) GCN-FP Accuracy 59.6% # 13
Node Classification Cora (3%) GCN-FP Accuracy 71.7% # 13
Node Classification Cora with Public Split: fixed 20 nodes per class GCN-FP Accuracy 74.6% # 31
Drug Discovery HIV dataset GraphConv AUC 0.822 # 2
Graph Regression Lipophilicity GC RMSE 0.655 # 3
Drug Discovery MUV GraphConv AUC 0.836 # 3
Drug Discovery PCBA GraphConv AUC 0.855 # 2
Node Classification PubMed (0.03%) GCN-FP Accuracy 56.2% # 11
Node Classification PubMed (0.05%) GCN-FP Accuracy 63.2% # 12
Node Classification PubMed (0.1%) GCN-FP Accuracy 70.3% # 12
Node Classification PubMed with Public Split: fixed 20 nodes per class GCN-FP Accuracy 76.0% # 27
Drug Discovery Tox21 GraphConv AUC 0.846 # 5
Drug Discovery ToxCast GraphConv AUC 0.754 # 3

Methods


No methods listed for this paper. Add relevant methods here