In recent years, breakthroughs in sequencing technology have made it easy to obtain sequence information of proteins and nucleic acids, providing momentum for data-driven research. Proteins are very important for various diseases and normal body activities, and their related sequence and structure information has been widely studied. The first step in conducting research is to determine how to digitally represent protein-related biological information. We categorize protein representation methods into three major types based on the properties of different representation methods: those based on inherent sequence properties, those based on physicochemical properties, and those based on structural properties. These characterization methods each have their unique features and advantages. By applying predefined computational rules, protein information is transformed into feature matrices of specific dimensions, enabling the operation of artificial intelligence algorithms. Moreover, selecting appropriate characterization methods for different use cases or combining these methods in a rational manner can significantly enhance model performance. Finally, several tools and platforms are introduced, such as PROFEAT, Scratch Protein Predictor, and POSSUM, which provide various protein representations and computational functions, allowing researchers to perform multidimensional analysis and prediction based on protein sequences.