Abstract
The Calpha positions for a protein can provide a scaffold for the reconstruction of more complete models. Reconstructions can be by manual rebuilding, from geometric solutions to the constraints on main-chain torsion angles or from databases of known protein structures. The last method is usually the most convenient and reliable. This paper describes a database reconstruction program, CALPHA, and assesses its accuracy and reliability by test reconstructions of well refined structures. Typically, backbone atoms are repositioned to within 0.3 A of their original positions. This corresponds to regenerating main-chain torsion angles to within 15 degrees. Uses of CALPHA for automating refinement procedures are discussed. In particular, the uses of Calpha-only and rcconstructed polyalanine models of HIV-1 reverse transcriptase for cross-rotation and translation function searches are compared. The CALPHA polyalanine model is found to provide more selectivity for approximately correct orientations. The effect on the translation function is dependent on the resolution shell employed. It is expected that these observations will be applicable in other cases