Flexible fitting of AlphaFold2-predicted models to cryo-EM density maps using elastic network models: a methodical...

Area of Science:

• Structural biology
• Computational biology
• Biophysics

Background:

• AlphaFold2 accurately predicts protein structures, but some models require refinement.
• Cryo-electron microscopy (cryo-EM) provides experimental structural data.
• Elastic Network Models (ENMs) describe protein flexibility.

Purpose of the Study:

• To develop and optimize a flexible refinement method for AlphaFold2 models using ENM normal modes.
• To improve the accuracy of AlphaFold2 models that deviate from experimental cryo-EM data.

Main Methods:

• Flexible refinement of AlphaFold2 models against cryo-EM maps.
• Utilizing normal modes from ENMs as basis functions for structural displacement.
• Systematic optimization of ENM motion model parameters.

Main Results:

• Successfully refined four AlphaFold2 models with improved TM-scores (e.g., respiratory supercomplex from 0.52 to 0.69).
• Identified optimal parameters including mode range (1-12), masked maps, inner product similarity, and Powell optimization.
• Integrated optimized parameters into the ModeHunter package (version 1.4).

Conclusions:

• Flexible refinement using ENM normal modes is effective for improving low-accuracy AlphaFold2 models.
• Optimized parameters enhance the performance of the refinement process.
• The ModeHunter package provides a practical tool for structural model refinement.