PSI Structural Biology Knowledgebase

PSI | Structural Biology Knowledgebase
Header Icons

Related Theme

Refining protein structure prediction

SBKB [doi:10.1038/sbkb.2011.68]
Technical Highlight - March 2012
Short description: Molecular dynamics simulations are shown to systematically improve template-based models of protein structure.

The large number of protein structures that have been experimentally solved makes it possible, when confronted with the sequence of a new protein, to make predictions about its structure. But doing this accurately and at atomic resolution remains challenging.

Superposition of the initial model (green), the refined model (blue) and the native structure (red) of TR614. Figure courtesy of Yang Zhang.

A standard approach to structure prediction involves the alignment of the query protein sequence to a template with known structure: in aligned regions, template structure then guides predictions for the query protein. Multi-template methods in particular can generate reasonable predictions, but the resulting models often have problematic local distortions. Zhang and colleagues now systematically test the use of molecular dynamics (MD) simulations, in which the positions of all atoms are modeled from fundamental physical principles, to refine template-based protein structure predictions.

The authors begin with a structural model for the query protein generated by the I-TASSER pipeline (a multiple template approach) and use models of either the full-length protein or of defined subfragments to search through the PDB. Global or local fragment templates identified in this way are used to generate distance maps that are then applied to constrain the MD simulation and refine the original model. The simulation employs the AMBER99 force field and also incorporates knowledge-based H-bonding and repulsive potentials. Zhang and colleagues name their approach Fragment-Guided Molecular Dynamics (FG-MD).

On a benchmark set of 181 proteins with known (and diverse) structures, FG-MD resulted in structural predictions that are closer to the native protein structure, based on several standard metrics. The greatest benefit accrued from local fragment restraints and the most substantial quality improvement was on models with strong starting structures. The authors propose that FG-MD serves to guide the simulations through a funnel-shaped intermediate energy landscape to a more native structure.

Finally, FG-MD was tested on datasets from Critical Assessment of protein Structure Prediction (CASP), retrospectively on 12 proteins from the CASP8 experiment and on a further 14 proteins as part of the CASP9 challenge. On these datasets as well, FG-MD consistently improved structure predictions and brought the models closer to the native structure.

Natalie de Souza


  1. Zhang J. et al. Atomic-level protein structure refinement using fragment-guided molecular dynamics conformation sampling.
    Structure 19, 1784-1795 (2011). doi:10.1016/j.str.2011.09.022

Structural Biology Knowledgebase ISSN: 1758-1338
Funded by a grant from the National Institute of General Medical Sciences of the National Institutes of Health