Abstract: The protein-protein interactions play an important role in life science. At present, many methods are developed with preferences of the amino acid residues, which do not offer the relative spatial information for the residue groups. However, the spatial information for the residue groups is important in the design of the protein-protein interactions. We proposed a new model, which is named ‘tri-prism’ model, by deep mining the existing protein-protein interaction patterns and refining the preference and the relative spatial information for the combination pairs of the residue groups. The model not only provided the preferences, but also offered the relative spatial information for the triplets combination pairs of the residue groups. The model was able to analyze the triplets combination pairs of the residue groups based on the preference factor, amino acid composition, and protein secondary structure. The model was applied to the interface of the PD-1/PD-L2 protein. According to the diversity characters of the composition and the spatial information between the combination pairs of the residue groups at the interface of the PD-1/PD-L2 protein and the predicted ones, we put forward the suggestions for the mutations of the residues, which offered a new view in the study of protein-protein interactions.