Can molecular dynamics simulations help in discriminating correct from erroneous protein 3D models?

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein structure prediction

Background:

• Current protein structure prediction methods often use simplified models, limiting accuracy.
• Physics-based energy functions and advanced sampling techniques like molecular dynamics (MD) simulations offer higher resolution modeling.
• Distinguishing correct protein models from incorrect ones is crucial for reliable structure prediction.

Purpose of the Study:

• To evaluate protein structural properties derived from MD simulations for discriminating correct from erroneous protein models.
• To assess the effectiveness of various criteria in identifying accurate sequence-structure alignments.

Main Methods:

• Generated protein models using the FROST fold recognition method, ranging from native structures to random alignments.
• Performed 11 nanoseconds (ns) of MD simulations at three temperatures for each model.
• Monitored Root-Mean-Square deviation (RMSd), RMSd fluctuations, conformational clustering, secondary structure evolution, and residue surface area along MD trajectories.

Main Results:

• No single criterion (RMSd, fluctuations, clustering, secondary structure, surface area) was fully effective in discriminating correct from erroneous models.
• RMSd, fluctuations, secondary structure, and clustering showed false positives; residue surface area showed false negatives.
• Combining multiple monitored criteria enabled straightforward discrimination between correct and erroneous models.

Conclusions:

• The combined analysis of structural properties from MD simulations significantly improves the ability to discriminate correct from erroneous protein models.
• This enhanced discrimination capability can refine the specificity and sensitivity of fold recognition methods, particularly for ambiguous cases.