Abstract: The disease caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is threatening people’s health and lives.Identifying phosphorylation sites is an important step in understanding the molecular mechanism of SARS-CoV-2.Due to the limitations of experimental methods,it is very necessary to establish effective prediction models.Therefore,a new SARS-CoV-2 phosphorylation site prediction model,Self-DeepIPs,is proposed.The protein sequence information is converted into digital information using dipeptide composition (DC),enhanced amino acid composition (EAAC),composition,transformation and distribution (CTD) and BLOSUM62.These features are also fused end-to-end,and the mutual information (MI) method is used to remove redundant information.The combination of BILSTM and the self-attention mechanism is used to build a deep learning model to predict the phosphorylation sites of the SARS-CoV-2.Then,five-fold cross-validation is used to test the model.The ACC and AUC values on the training set reach 83.62% and 91.70% respectively,and the ACC and AUC values on the independent test set reach 82.56% and 91.23% respectively.The experimental results show that the Self-DeepIPs method proposed in this paper can effectively identify SARS-CoV-2 phosphorylation sites.