Table 5 Imputation performance of GNNImpute and GCN architecture model (Klein dataset)
From: An efficient scRNA-seq dropout imputation method using graph attention network
MSE | MAE | PCC | CS | ARI | NMI | |
|---|---|---|---|---|---|---|
GCN (without attention) | 6.7994 | 0.8841 | 0.8805 | 0.8882 | 0.7998 | 0.8049 |
GNNImpute (1 attention head) | 5.7850 | 0.9059 | 0.8779 | 0.8867 | 0.6945 | 0.7232 |
GNNImpute (3 attention heads) | 6.1119 | 0.8725 | 0.8867 | 0.8942 | 0.8183 | 0.8204 |
GNNImpute (5 attention heads) | 4.9420 | 0.8629 | 0.8950 | 0.9024 | 0.8155 | 0.8311 |
GNNImpute (8 attention heads) | 6.4669 | 0.8592 | 0.8921 | 0.8988 | 0.8202 | 0.8366 |