Predictions of Protein-Protein Interfaces within Membrane Protein Complexes

Background: Prediction of interaction sites within the membrane protein complexes using the sequence data is of a great importance, because it would find applications in modification of molecules transport through membrane, signaling pathways and drug targets of many diseases. Nevertheless, it has gained little attention from the protein structural bioinformatics community. Methods: In this study, a wide variety of prediction and classification tools were applied to distinguish the residues at the interfaces of membrane proteins from those not in the interfaces. Results: The tuned SVM model achieved the high accuracy of 86.95% and the AUC of 0.812 which outperforms the results of the only previous similar study. Nevertheless, prediction performances obtained using most employed models cannot be used in applied fields and needs more effort to improve. Conclusion: Considering the variety of the applied tools in this study, the present investigation could be a good starting point to develop more efficient tools to predict the membrane protein interaction site residues.