Accurate flexible refinement for atomic-level protein structure using cryo-EM density maps and deep learning.

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Accurate protein structure prediction is crucial for understanding biological function.
• Deep learning advancements in cryo-electron microscopy (cryo-EM) and protein structure prediction necessitate improved refinement methods.
• Integrating cryo-EM density maps with predicted contact/distance maps offers a promising avenue for enhancing protein structure models.

Purpose of the Study:

• To develop and validate a novel protein structure optimization strategy.
• To improve the accuracy of protein structure refinement using cryo-EM density maps and deep learning-predicted contact/distance maps.
• To establish a robust method for atomic-level structure refinement.

Main Methods:

• A deep learning-based protein structure optimization method was proposed.
• Physics- and knowledge-based energy functions were integrated with cryo-EM density map and deep learning data.
• A dynamic confidence score was introduced to guide the iterative refinement process, balancing density map and contact/distance map influence.
• The protocol was tested on 224 non-homologous membrane proteins.

Main Results:

• The method successfully generated 214 structural models with correct folds from 224 tested proteins.
• A small percentage (4.5%) of correct folds were achieved even when starting from models with incorrect folds.
• The proposed method demonstrated superior performance compared to other state-of-the-art techniques.
• Key advantages include effective utilization of density and contact/distance maps and a novel energy function for re-assembly.

Conclusions:

• The developed strategy represents a valuable and ready-to-use approach for atomic-level protein structure refinement.
• The integration of cryo-EM density maps and predicted contact/distance maps via deep learning significantly improves structure refinement accuracy.
• This method holds potential for advancing structural biology research by providing more reliable protein models.