High-quality protein backbone reconstruction from alpha carbons using Gaussian mixture models.

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein modeling

Background:

• Coarse-grained protein models accelerate structural modeling but require accurate full-atom reconstruction.
• Existing methods for reverting to all-atom structures from coarse-grained models have limitations in speed and accuracy.

Purpose of the Study:

• To present a novel algorithm (PD2) for reconstructing main-chain protein structures from C-alpha traces.
• To evaluate the accuracy and performance of the PD2 method against existing approaches.

Main Methods:

• Parameterization of the PD2 algorithm using Gaussian mixture models (GMMs) fitted to backbone fragments.
• Statistical comparison of PD2 with other methods using RMSD and dihedral angle differences.
• Assessment of Ramachandran angle distributions and Phaser molecular replacement log-likelihood gains.
• Correlation analysis between backbone reconstruction accuracy and sidechain reconstruction using SCWRL4.
• Evaluation of full-atom model energies using Rosetta.

Main Results:

• The PD2 method demonstrates statistically significant improvements in accuracy over competing methods.
• PD2 achieves more realistic Ramachandran dihedral angle distributions compared to real proteins.
• Sidechain reconstruction accuracy shows a significant correlation with backbone reconstruction accuracy.
• PD2-generated models result in significantly lower energies when refined with Rosetta.

Conclusions:

• The PD2 algorithm offers a fast and accurate solution for reconstructing full-atom protein models from coarse-grained representations.
• The method has positive implications for multiscale protein modeling strategies.
• PD2 is available as a webserver and C++ source code for non-commercial use.