Computational prediction of protein-protein interaction (PPI) from protein sequence is important as many cellular functions are made possible through PPI. The Protein Interaction Prediction Engine (PIPE) software suite was developed at Carleton University for such predictions. This thesis aims to conduct a thorough performance assessment of the PIPE-Sites predictor through the use of a large high-quality set of known PPI sites. The results show that PIPE-Sites has relatively low accuracy even after retuning the inherent hyperparameters of the method. Furthermore, PIPE-Sites are shown to be ineffective when applied to similarity-weighted score data. Thus, three new sequence-based methods of predicting PPI sites are proposed and evaluated, including the Panorama, BrightSpot, and ClusterNet methods. The new methods leverage similarity-weighted score data to further increase performance. Ultimately, ClusterNet significantly outperforms the other methods over two different performance metrics when evaluated on both human and yeast data PPI site data.