Reoptimized UNRES Potential for Protein Model Quality Assessment.

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Accurate protein structure prediction is crucial for understanding biological function.
• Evaluating the quality of protein models remains a significant challenge in bioinformatics.
• Existing methods often struggle to reliably rank models based on native structure similarity and folding pathways.

Purpose of the Study:

• To reoptimize the UNited RESidue (UNRES) force field for improved protein structure model selection.
• To assess the performance of a reoptimized UNRES (OUNRES) using deep neural networks and enhanced features.
• To evaluate OUNRES's effectiveness in ranking models from the CASP11 dataset and tracking protein unwinding.

Main Methods:

• Simulated the use of the UNRES force field on CASP11 server data.
• Reoptimized UNRES using a deep feed-forward neural network with additional descriptive features.
• Compared the performance of the original UNRES and the reoptimized OUNRES in model selection and scoring.

Main Results:

• The reoptimized UNRES (OUNRES) demonstrated superior performance in selecting the best protein structure models.
• OUNRES showed improved accuracy in tracking protein unwinding from its native state.
• The reoptimization procedure significantly improved the correlation between UNRES values and the Template Modeling Score (TMS).

Conclusions:

• Reoptimized UNRES (OUNRES) offers a more effective tool for ranking protein structure models.
• The integration of deep learning and enhanced features in force field reoptimization yields better predictive power.
• OUNRES provides a valuable advancement for protein structure prediction and analysis.