Predicting pHLA Binding Affinity Using CNN with Step Connections

Prediction of peptide-HLA binding affinity is a crucial step in cancer immunotherapy targeting neoantigens. Currently, machine learning algorithms, particularly neural networks such as convolutional neural networks (CNNs), are primarily used for pHLA binding affinity prediction. However, these network architectures do not incorporate the latest advancements in deep learning, leaving room for further performance improvement. By adding skip connections in CNNs, the information flow through the network can be strengthened, leading to enhanced overall network efficiency and improved model performance. It should be noted that adding skip connections introduces additional parameters, increasing model capacity and making it more prone to overfitting. This issue can be mitigated by appropriately increasing the dropout rate in the Dropout layer. Experimental results demonstrate that CNNs with skip connections show promising performance in pHLA binding affinity prediction.