Computational prediction of protein-protein interactions in sphingolipid signaling network

Abstract

Proteins carry out most of the work in the cell such as immunological recognition, DNA repair and replication, enzymatic activity, cell signaling by interacting with other proteins. Therefore deciphering protein-protein interactions can give useful information about various mechanisms in the cell and hence provide important clues on disease states such as cancer and apoptosis. In this work, protein-protein interactions involved in sphingolipid metabolism were investigated in an effort to elucidate sphingolipid metabolism. Three categories of proteins were examined; proteins with previously identified partners, proteins with no known partners and clusters (Cluster A, Cluster B and Cluster C). First, the missing interaction partners of six proteins with no known interactions were identified by sequence based prediction methods and then filtered using the GO annotations of protein partners. The putative interaction partners were determined as; YHR135C with YDR294C; YHL020C with YER019W; YKL126W with YGR143W; YPL204W and YHR135C with YGR212W; YAR033W and YGL051W with YJL134W; YAR033W and YGL051W with YKR053C. The structures for these proteins were predicted by homology modeling and the structures of protein complexes were predicted by protein-protein docking. Nearly half of the complexes of the proteins with predicted partners formed biological contacts which mean these model interactions may occur in real systems. Next, the hotspots in every model of interacting protein pairs were identified by KFC. The model with the maximum number of closest hotspots was selected as the putative model structure for that protein complex. The hotspots were then classified according to their chemical features such as acidity, polarity, hydrophobicity and enrichment of certain amino acids. More than 60% of the hotspot residues in all categories of protein complexes are hydrophobic. The most repeated hotspot residue was found to be TYR in Clusters A and B, whereas it was LEU (15.04%) in Cluster C and in protein complexes with known interaction partners.