A descriptive analysis of populations of three-dimensional structures calculated from primary sequences of proteins by OSIRIS

Among different ab initio approaches to calculate 3D-structures of proteins out of primary sequences, a few are using restricted dihedral spaces and empirical equations of energy as is OSIRIS. All those approaches were calibrated on a few proteins or fragments of proteins. To optimize the calculation over a larger diversity of structures, we need first to define for each sequence what are good conditions of calculations in order to choose a consensus procedure fitting most 3D-structures best. This requires objective classification of calculated 3D-structures. In this work, populations of avian and bovine pancreatic polypeptides (APP, BPP) and of calcium-binding protein (CaBP) are obtained by varying the rate of the angular dynamics of the second step of OSIRIS. Then, 3D-structures are clustered using a nonhierarchical method, SICLA, using rmsd as a distance parameter. A good clustering was obtained for four subpopulations of APP, BPP and CaBP. Each subpopulation was characterized by its barycenter, relative frequency and dispersion. For the three alpha-helix proteins, after the step I of OSIRIS, most secondary structures were correct but molecules have a few atomic contacts. Step 2, i.e., the angular dynamics, resolves those atomic contacts and clustering demonstrates that it generates subpopulations of topological conformers as the barycenter topologies show.