Protein-protein Predictions (PPI) are crucial to understanding the biological functions in living organisms. State-of- the-art computational techniques to predict PPIs use Supervised Learning, which requires a representative dataset of already known PPIs for the organism under study. However, there exist scenarios where few PPIs are identified for an organism, causing insufficient training data for learning a computational model. Our goal in this paper is to tackle the problem of lack of representative dataset and leverage the current known PPIs in other organisms to predict the PPIs in the organism under study. The methodology used to address the issue, is based on transfer learning approach. For evaluation purpose, we will use sets of two species with known PPIs and provide the accuracy of our algorithm. The possibility of cross-species training and prediction can lead us to the discovery of novel PPIs for understudy organisms and help save resources spend on wet-lab experiments.