Table 3.

Generation Performances of different network architectures

Model	high-affinity rate
2CNN + GAP	0.756
1CNN + GAP	0.813
2CNN + muti-GAP	0.859
3CNN + GAP	0.481
random data	0.05

The dataset is HLA-A*0201. Except for the random data model, all the model is a variety of MAM network. High-affinity rate denotes the fraction of peptides with high affinity in all the generated peptide. Random data is to create data randomly at all sites. GAP stands for global averaging pooling. “A CNN + B GAP” represents A numbers of CNN layer and B numbers of Global Pooling Layer in the feature extraction part. All the definitions we mentioned are the same