Immune checkpoint proteins (ICPs) are key regulators of immune responses and serve as crucial therapeutic targets in immuno-oncology. In this study, we present a computational framework for ICP identification based on a hybrid deep learning architecture that integrates Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory networks (BiLSTM). This architecture is specifically designed to process long protein sequences, capturing both local motifs and long-range contextual dependencies. To enhance feature representation, we incorporate Evolutionary Scale Modeling (ESM), a pretrained protein language model that encodes rich semantic and structural information from amino acid sequences. Experimental results demonstrate that the CNN + BiLSTM model combined with ESM features achieves high predictive performance, with an accuracy of 0.9807, MCC of 0.9499, sensitivity of 0.9912, specificity of 0.9745, and an AUC of 0.9986 on an independent test set. These results highlight the potential of integrating protein language models with deep learning for sequence-based protein classification, offering a scalable and effective solution for functional protein annotation and the discovery of novel immunotherapeutic targets.