Modeling experimental image formation for likelihood-based classification of electron microscopy data

Area of Science:

• Structural biology
• Biophysics
• Computational biology

Background:

• Three-dimensional cryo-electron microscopy (3D cryo-EM) is challenged by structural heterogeneity in large macromolecular complexes.
• Maximum likelihood (ML) approaches offer a promising solution for image classification in such complex samples.

Purpose of the Study:

• To develop a statistical data model for 2D and 3D maximum likelihood refinement in cryo-EM.
• To improve the alignment and classification of structurally heterogeneous projection data.

Main Methods:

• A statistical data model was formulated to describe experimental image formation in Fourier space.
• The model incorporates a spatial frequency-dependent noise model and defocus-dependent imaging effects.
• Expectation-Maximization-like algorithms were developed for data alignment and classification.

Main Results:

• The proposed statistical model and algorithms were applied to 2D classification of archaeal helicase MCM.
• Effectiveness was demonstrated through 3D classification of the 70S E. coli ribosome.
• Successful 3D classification was also achieved for Simian Virus 40 large T-antigen projections.

Conclusions:

• The developed statistical data model and ML refinement algorithms effectively address structural heterogeneity in cryo-EM.
• This approach enhances the capability of cryo-EM for analyzing large, complex macromolecular machines.
• The methods are broadly applicable to diverse projection datasets in structural biology.