Table 5 Performance of adaptor protein prediction on independent testing set using different numbers of hidden nodes in transformer. The model has three convolutional blocks with 20 convolution kernel
From: A lightweight classification of adaptor proteins using transformer networks
Hidden nodes | AUC | MCC | Model size |
|---|---|---|---|
20 | 0.8941 | 0.4668 | 7.7k |
32 | 0.9042 | 0.4696 | 8.2k |
64 | 0.8978 | 0.456 | 9.5k |
80 | 0.8998 | 0.4646 | 10.1k |
128 | 0.9031 | 0.4872 | 12.1k |
200 | 0.8999 | 0.4639 | 15.0k |
256 | 0.8978 | 0.4633 | 17.3k |