Table 2.

Binding Affinity Prediction Performances of different network architectures

Model	SRCC	AUC
2CNN+FC	0.178	0.56
2CNN + GAP	0.083	0.554
1CNN + GAP	0.119	0.575
2CNN + muti-GAP	0.117	0.576
3CNN + GAP	0.139	0.59

The training dataset is HLA-A*0201 while the test dataset is IEDB 1029824 HLA-A*0201 segmented from HLA-A*0201. FC denotes full-connected layer. SRCC stands for Spearman’s rank correlation coefficient and AUC stands for area under the receiver operating characteristic curve. All the models are well-trained. “A CNN+B GAP” represents A CNN layers and B Global Pooling Layers in the feature caught part. The “2CNN + muti-GAP” is our final MHC-CNN predictor