Protein-protein interactions (PPIs) are fundamental to the functioning of biological systems, regulating a myriad of cellular processes. Therefore, accurate prediction of PPIs is crucial for disease diagnosis and the development of novel therapeutic interventions. Computational techniques streamline the forecasting of PPIs and prove to be more cost-effective in terms of both resources and time compared to experimental approaches. To date, the majority of research on PPI has predominantly concentrated on sequence information. In this study, graph neural network (GNN)-based GraphSAGE model, called GSPPI, has been leveraged to predict the interactions between proteins by utilizing graphlet features (up to five nodes) which is a node’s local network topological property. The benchmark positive dataset has been curated from the HIPPIE repository. The overall accuracy, precision, and recall of 0.93, 0.91, and 0.95 have been obtained in this proposed work on holdout set for fivefold cross-validation (cv). In addition, Graph Convolutional Network (GCN)- and Graph Attention Network (GAT)-based GNNs have been implemented, and the results are compared with GSPPI on the same dataset to predict PPIs. The results acquired affirm the effectiveness of the suggested method, surmounting the performance of the previously dominant techniques.