Neural Message Passing for Quantum Chemistry

ICML 2017 Justin GilmerSamuel S. SchoenholzPatrick F. RileyOriol VinyalsGeorge E. Dahl

Supervised learning on molecules has incredible potential to be useful in chemistry, drug discovery, and materials science. Luckily, several promising and closely related neural network models invariant to molecular symmetries have already been described in the literature... (read more)

PDF Abstract

Evaluation results from the paper


Task Dataset Model Metric name Metric value Global rank Compare
Node Classification CiteSeer (0.5%) MPNN Accuracy 41.8% # 12
Node Classification CiteSeer (1%) MPNN Accuracy 54.3% # 11
Node Classification CiteSeer with Public Split: fixed 20 nodes per class MPNN Accuracy 64.0% # 15
Node Classification Cora (0.5%) MPNN Accuracy 46.5% # 11
Node Classification Cora (1%) MPNN Accuracy 56.7% # 11
Node Classification Cora (3%) MPNN Accuracy 72.0% # 10
Node Classification Cora with Public Split: fixed 20 nodes per class MPNN Accuracy 78.0% # 13
Graph Regression Lipophilicity MPNN RMSE 0.719 # 4
Node Classification PubMed (0.03%) MPNN Accuracy 53.9% # 11
Node Classification PubMed (0.05%) MPNN Accuracy 59.6% # 11
Node Classification PubMed (0.1%) MPNN Accuracy 67.3% # 11
Node Classification PubMed with Public Split: fixed 20 nodes per class MPNN Accuracy 75.6% # 14
Drug Discovery QM9 MPNNs Error ratio 0.68 # 1
Formation Energy QM9 MPNN MAE 0.45 # 2