Table 3 Performance of different modules on the BindingDB regression dataset
Modules | Methods | MSE \(\downarrow\) | Pearson \(\uparrow\) | \(r^2_m\) \(\uparrow\) |
|---|---|---|---|---|
Protein embedding | one-hot | \(0.578\pm 0.003\) | \(0.856\pm 0.000\) | \(0.734 \pm 0.002\) |
ProtVec | \(0.496\pm 0.003\) | \(0.879\pm 0.000\) | \(0.773 \pm 0.001\) | |
TAPE | \({\varvec{0.474\pm 0.003}}\) | \(0.884\pm 0.001\) | \(0.782 \pm 0.001\) | |
ESM | \({\varvec{0.474\pm 0.003}}\) | \({\varvec{0.885\pm 0.001}}\) | \({\varvec{0.783 \pm 0.001}}\) | |
Feature encoder | GCN-RNN | \(0.485\pm 0.002\) | \(0.882\pm 0.001\) | \(0.778 \pm 0.001\) |
GCN-GRU | \(0.490\pm 0.003\) | \(0.880 \pm 0.001\) | \(0.775 \pm 0.001\) | |
GCN-LSTM | \(0.485\pm 0.004\) | \(0.883\pm 0.001\) | \(0.780 \pm 0.002\) | |
GraphSAGE-RNN | \(0.477\pm 0.003\) | \(0.884\pm 0.001\) | \(0.782 \pm 0.002\) | |
GraphSAGE-GRU | \(0.477\pm 0.004\) | \(0.884\pm 0.001\) | \(0.782 \pm 0.002\) | |
GraphSAGE-LSTM | \({\varvec{0.474\pm 0.003}}\) | \({\varvec{0.884\pm 0.001}}\) | \({\varvec{0.782 \pm 0.001}}\) | |
Joint module | Concatenate | \(0.570\pm 0.004\) | \(0.859\pm 0.001\) | \(0.738 \pm 0.002\) |
Kronecker product | \({\varvec{0.474\pm 0.003}}\) | \({\varvec{0.884\pm 0.001}}\) | \({\varvec{0.782 \pm 0.001}}\) |