Cross-validation in cryo-EM-based structural modeling.

Area of Science:

• Structural Biology
• Biophysics
• Biochemistry

Background:

• Single-particle cryo-electron microscopy (cryo-EM) determines macromolecular structures at subnanometer resolution.
• Atomic model refinement into cryo-EM density maps is common but prone to overfitting due to lower resolution compared to X-ray crystallography.
• Overfitting and density misinterpretation are significant challenges in cryo-EM structure determination.

Purpose of the Study:

• To introduce a novel cross-validation approach for real-space refinement against cryo-EM density maps.
• To address the overfitting problem prevalent in cryo-EM structure refinement.
• To provide a method for optimizing refinement restraints and assessing model accuracy.

Main Methods:

• Developed a cross-validation strategy analogous to crystallography for real-space refinement.
• Applied the method to simulated and experimental cryo-EM data, including the rotavirus double-layer particle.
• Introduced a method to quantify correlations between structure factor sets for dataset splitting.

Main Results:

• The proposed cross-validation approach effectively detects overfitting in cryo-EM density map refinement.
• The method aids in optimizing the selection of restraints during the refinement process.
• Demonstrated successful application on multiple protein structures and experimental data.

Conclusions:

• The developed cross-validation technique is crucial for reliable atomic model building in cryo-EM.
• This method enhances the accuracy and trustworthiness of structures determined by cryo-EM.
• The analysis of structure factor correlations has broader applications in structural biology and cryo-EM data processing.