An efficient method for reconstructing protein backbones from alpha-carbon coordinates

We present an approach for building protein backbones from alpha-carbon (Calpha) coordinates. The approach is analytical and based on the information of favored regions in the Ramachandran map. The backbone construction consists of three parts: prediction of (phi, psi) angle pairs from the Calpha trace, generation of atomic coordinates with these (phi, psi) angles, and refinement by subsequent energy minimization. Tests on several known protein structures show that the root mean square deviations in reconstructed backbones are 0.25-0.48 A for coordinates and 14-34 degrees for phi and psi angles. The results indicate that our method is one of the best methods proposed in terms of accuracy. It has also been revealed that the approach is not only robust against errors in Calpha coordinates but is also capable of providing equivalent or more reasonable models compared to other known methods. Furthermore, backbone structures were found to be built accurately by using the (phi, psi) angles from a different structure of the same protein. This suggests that the approach could be effective and useful in homology modeling studies.