Table 2 Average test scores (± standard deviation) obtained by the different models in the protein hold out use case (UC2). The suffix +emb indicates that protein sequence embeddings were concatenated to the I/O features. Best results are boldfaced
From: Sensitivity analysis on protein-protein interaction networks through deep graph networks
Model | MCC | F1 | ACC | AUROC |
|---|---|---|---|---|
Null | \(.000{\scriptstyle \pm .000}\) | \(.000{\scriptstyle \pm .000}\) | \(.680{\scriptstyle \pm .024}\) | \(.500{\scriptstyle \pm .000}\) |
DeepSets | \(.021\scriptstyle \pm .036\) | \(.218\scriptstyle \pm .200\) | \(.572\scriptstyle \pm .169\) | \(.560\scriptstyle \pm .142\) |
DeepSets+emb | \(.415\scriptstyle \pm .031\) | \(.605\scriptstyle \pm .038\) | \(.741\scriptstyle \pm .013\) | \(.783\scriptstyle \pm .033\) |
DGN | \(.404\scriptstyle \pm .077\) | \(.604\scriptstyle \pm .043\) | \(.728\scriptstyle \pm .052\) | \(.791\scriptstyle \pm .038\) |
DGN+emb | \(\mathbf {.512\scriptstyle \pm .040}\) | \(\mathbf {.667\scriptstyle \pm .032}\) | \(\mathbf {.788\scriptstyle \pm .019}\) | \(\mathbf {.844\scriptstyle \pm .025}\) |