Incorporating constraints from low resolution density map in ab initio structure prediction using Rosetta

Abstract

We have developed a new method for adding constraints derived from low resolution density maps to Rosetta ab initio prediction method. This method incorporates the geometrical constraints of the helix skeleton that can be detected from a low resolution density map. We propose a 2-stage approach to predict the backbone of a protein from a low resolution map. In stage one, a small set of possible topologies will be predicted for the helix skeleton [1]. This paper describes the second stage that is to predict the backbone of the protein from a low resolution density map. A constraint scoring function was developed and incorporated in the Rosetta simulation process. The entire density map is only used for the final selection among the possible backbones that satisfy the constraints. Our method was tested with 16 mainly alpha-helical proteins ranging from 50 to 150 residues. 12 of the 16 proteins show improved accuracy for both the top 1 prediction and the best of the top 5 predictions. The average improvement of the RMSD to native is 4.76 A for the top 1 model and 3.05 A for the best of the top 5 ranked models when the density map is applied.