Abstract: To address the current issue of low performance in predicting the binding affinity between antigenic peptides and specific MHC class II molecules, which fails to meet clinical requirements, we proposed T5MHCII, a deep learning-based prediction model for the affinity of MHC II class molecules to peptides. The model employed the knowledge previously acquired from the protein language model ProtT5 to extract the amino acid sequences via a transfer learning approach, thereby generating high-quality characterizations. This knowledge was then integrated with the robust learning abilities of deep learning to develop a novel model with enhanced predictive capabilities. The results of the five-fold cross-validation demonstrated that the model exhibited superior performance compared to NetMHCIIpan-3.2, PUFFIN, DeepMHCII, and RPEMH, with an AUC of 0.893±0.003 and a PCC of 0.780±0.006. The leave-one-out cross-validation (LOOCV) further demonstrated that the model exhibited enhanced generalization capabilities. This study proposes a novel approach to enhance the precision of peptide-MHCII prediction in the context of limited data affinity through the application of deep learning techniques.