Skip to main content

Table 1 Architecture of MHC-CNN network

From: In silico design of MHC class I high binding affinity peptides through motifs activation map

Type

Notes

Input layer

 

Embedding(each site vec dim = 15)

Finally build N*15 matrix(N is the mer number in peptide, N is 9 denotes 9 mer)

Conv1D[filter_size=16, filter_length=7] + LeakyReLU(0.3)

Low-level feature

Dropout(0.25)

 

Conv1D[filter_size=32, filter_length=7] + LeakyReLU(0.3)

High-level feature

Dense layer1(1) without bias

Global averaging Pooling network, input is the first Conv1D

Dense layer2(1) without bias

Global averaging Pooling network, input is the second Conv1D

Dense layer3(1) without bias

Fusion of the different level GAP layers(aka voting method)

Sigmoid [prediction]

Â