Table 6 Results of the Variants of the Proposed Model for Molecular Subtype Classification Using the TCGA Pan-cancer Dataset
From: A multimodal graph neural network framework for cancer molecular subtype classification
GNN layers (Module) | 300 | 500 | 700 | |||
|---|---|---|---|---|---|---|
Accu.a | F1 | Accu.a | F1 | Accu.a | F1 | |
GAT (No Decoder) | 76.3% ± 1.6% | 0.76 ± 0.03 | 78.2% ± 1.2% | 0.77 ± 0.01 | 80.2% ± 1.2% | 0.79 ± 0.01 |
GCN (No Decoder) | 75.3% ± 1.2% | 0.74 ± 0.02 | 76.8% ± 0.8% | 0.75 ± 0.01 | 79.3% ± 0.8% | 0.78 ± 0.01 |
GAT (No Parallel) | 75.4% ± 1.8% | 0.73 ± 0.03 | 76.1% ± 1.7% | 0.73 ± 0.02 | 79.8% ± 1.3% | 0.78 ± 0.02 |
GCN (No Parallel) | 73.5% ± 1.2% | 0.72 ± 0.02 | 75.4% ± 1.2% | 0.73 ± 0.01 | 76.7% ± 0.8% | 0.75 ± 0.01 |
GAT (No Decoder & Parallel) | 74.9% ± 1.4% | 0.73 ± 0.02 | 76.4% ± 0.9% | 0.74 ± 0.01 | 80.1% ± 0.8% | 0.79 ± 0.01 |
GCN (No Decoder & Parallel) | 73.1% ± 1.2% | 0.73 ± 0.02 | 75.6% ± 0.8% | 0.73 ± 0.01 | 77.3% ± 0.02% | 0.76 ± 0.01 |