Protein-protein interactions play a crucial role in biological processes such as cell signal transduction, gene expression and metabolic regulation, and thus their identification is essential for understanding these complex biological processes. Predicting protein-protein interactions is a hot topic of great significance, which can provide assistances in areas such as drug discovery and protein function research and design as well. In recent years, with the development of artificial intelligence, machine learning technologies have been applied gradually to the prediction of protein-protein interactions, which has shown good potentials. However, when processing a large amount of protein information, traditional machine learning methods are difficult to mine the intrinsic patterns and potential features, and deep learning techniques are needed. Compared with the three-dimensional structure of proteins, sequence information is easier to obtain, and the development of high-throughput sequencing technology provides abundant protein sequence information, which greatly facilitates the development of sequence-based deep learning technologies. Sequence-based deep learning models predict protein-protein interactions by learning intrinsic patterns and features from protein sequence information, which greatly improves prediction efficiency and accuracy. In this review, we focus on progress of deep learning in predicting sequence-based protein interactions, categorize, which is summarized according to the algorithmic framework and timeline, briefly describing the construction methods of datasets and the evaluation metrics of the models, discussing in detail the sequence encoding methods and common algorithmic architectures, and demonstrating the computational models based on various types of algorithms and their features and advantages. Finally, we analyze current challenges in predicting protein-protein interactions using deep learning methods, and discuss possible solutions. With the development of deep learning technology, the efficiency of predicting protein-protein interactions has increased dramatically. As a result, there is a need to develop models with stronger generalization and more robust prediction capabilities to aid the prediction of protein-protein interactions in the future.