Table 1 Test set predictive performance
From: Explaining compound activity predictions with a substructure-aware loss for graph neural networks
Avg. RMSE | W. Avg. RMSE | Avg. PCC | W. Avg. PCC | |
|---|---|---|---|---|
RF | \(0.35\, (\pm 0.11)\) | \(0.30\, (\pm 0.08)\) | \(0.95\, (\pm 0.07)\) | \(0.96\, (\pm 0.04)\) |
GNN \({\mathcal {L}}_{\text{MSE}}\) | \(0.34\, (\pm 0.23)\) | \(0.25\, (\pm 0.13)\) | \(0.89\, (\pm 0.23)\) | \(0.96\, (\pm 0.08)\) |
GNN \({\mathcal {L}}_{\mathrm {MSE+AC}}\) | \(0.31\, (\pm 0.24)\) | \(0.24\, (\pm 0.13)\) | \(0.89\, (\pm 0.23)\) | \(0.96\, (\pm 0.07)\) |
GNN \({\mathcal {L}}_{\mathrm {MSE+UCN}}\) | \(0.47\, (\pm 0.28)\) | \(0.37\, (\pm 0.14)\) | \(0.84\, (\pm 0.24)\) | \(0.93\, (\pm 0.08)\) |