Bayesian inference of initial models in cryo-electron microscopy using pseudo-atoms.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Single-particle cryo-electron microscopy (cryo-EM) is vital for determining macromolecular assembly structures.
• Accurate initial 3D models and image orientations are critical for successful cryo-EM structure determination.
• Estimating these initial parameters presents a significant computational challenge due to numerous unknowns.

Purpose of the Study:

• To introduce a novel probabilistic algorithm for efficient initial model and orientation estimation in cryo-EM.
• To develop a fast algorithm with minimal parameters that provides precision information for estimated model parameters.
• To enhance the initial steps of cryo-EM data processing for improved 3D structure inference.

Main Methods:

• Utilized a pseudo-atomic model with isotropic Gaussian components for representing low-resolution 3D structures.
• Employed a Bayesian framework to jointly estimate 3D structure and image orientations.
• Applied Markov chain Monte Carlo (MCMC) sampling for parameter estimation, yielding an ensemble of models.

Main Results:

• The algorithm demonstrates speed and requires few parameters, simplifying the cryo-EM workflow.
• It effectively reduces the parameter space for 3D structure representation and projection computation.
• The MCMC sampling provides an ensemble of models, offering insights into parameter precision.

Conclusions:

• The developed probabilistic algorithm offers a robust and efficient solution for initial model and orientation estimation in cryo-EM.
• Its ability to provide parameter precision is valuable for assessing the reliability of the inferred structures.
• The method is validated on both simulated and real cryo-EM datasets, highlighting its practical applicability.