Fig. 2
From: In silico design of MHC class I high binding affinity peptides through motifs activation map
MHC-CNN predictor (The prediction part of our MAM Network). Two 1-D convolution layers are used to extract the hidden features. Global average pooling layer is to replace fully-connected layer and calculate the weights of every feature. Then two dense layer is to merge the features from two levels into one final binding score. The input of our predictor is peptide’s representation matrix while the output is the binding probability. Here we take 9-mer as an example