Predicting function from structure: examples of the serine protease inhibitor canonical loop conformation found in extracellular proteins

The prediction of protein function from structure is becoming of growing importance in the age of structural genomics. We have focused on the problem of identifying sites of potential serine protease inhibitor interactions on the surface of proteins of known structure. Given that there is no sequence conservation within canonical loops from different inhibitor families we first compare representative loops to all fragments of equal length among proteins of known structure by calculating main-chain RMS deviation. Fragments with RMS deviation below a certain threshold (hits) are removed if residues have solvent accessibilities appreciably lower than those observed in the search structure. These remaining hits are further filtered to remove those occurring largely within secondary structure elements. Likely functional significance is restricted further by considering only extracellular protein domains. Also a test is performed to see if the loop can dock into the binding site of the serine protease trypsin without unacceptable steric clashes. By comparing different canonical loop structures to the protein structure database we show that the method was able to detect previously known inhibitors. In addition, we discuss potentially new canonical loop structures found in secreted hydrolases, toxins, viral proteins, cytokines and other proteins. We discuss the possible functional significance of several of the examples found.