Introducing robustness to maximum-likelihood refinement of electron-microscopy data

Area of Science:

• Structural biology
• Biophysics
• Computational biology

Background:

• Electron microscopy (EM) is crucial for visualizing molecular structures.
• Maximum-likelihood refinement is key for accurate EM image analysis.
• Conventional methods often struggle with data outliers.

Purpose of the Study:

• To introduce a novel expectation-maximization (EM) algorithm for EM image refinement.
• To enhance robustness against atypical observations in EM datasets.
• To improve the accuracy of 3D classifications of biological macromolecules.

Main Methods:

• Developed an EM algorithm based on fitting mixtures of multivariate t-distributions.
• Applied the algorithm to experimental electron microscopy datasets.
• Included artificially generated outliers to test robustness.
• Compared results with conventional Gaussian mixture models.

Main Results:

• The t-distribution-based EM algorithm demonstrated robustness against outliers.
• Minor differences observed in 2D classifications compared to conventional methods.
• Significantly stronger elongation factor G density in 3D classification of heterogeneous ribosomes.

Conclusions:

• The novel algorithm offers improved performance for 3D classification in cryo-EM.
• Its robustness makes it suitable for datasets with inherent noise or artifacts.
• This method advances the analysis of structurally complex biological samples.