Confidence-guided cryo-EM map optimisation with LocScale-2.0

Area of Science:

• Structural biology
• Biophysics
• Computational biology

Background:

• Cryo-electron microscopy (cryo-EM) maps often exhibit variable quality, with high-resolution and low-resolution regions present simultaneously.
• This heterogeneity complicates the structural interpretation of macromolecular assemblies, particularly those that are structurally diverse.

Purpose of the Study:

• To develop a novel framework for optimizing cryo-EM maps without prior structural knowledge.
• To introduce a method for enhancing local details and preserving weak structural information in cryo-EM density maps.
• To provide a statistically grounded measure of reliability for interpreted regions in cryo-EM maps.

Main Methods:

• Development of LocScale-2.0, a context-aware map optimization framework.
• Introduction of LocScale-FEM, a Bayesian approximate deep-learning approach to emulate LocScale-2.0 optimization.
• Generation of feature-enhanced maps with voxel-wise confidence scores.

Main Results:

• LocScale-2.0 successfully enhances local detail and connectivity while preserving weak structural context.
• LocScale-FEM generates feature-enhanced maps with associated confidence scores.
• Confidence scores are sensitive to local phase error and indicate regions requiring cautious interpretation.

Conclusions:

• Confidence-guided map optimization aids biological interpretation in cryo-EM.
• The developed methods increase objectivity in the analysis of cryo-EM density data.
• LocScale-2.0 and LocScale-FEM offer valuable tools for structural biologists working with heterogeneous samples.