Deep learning-based mixed-dimensional Gaussian mixture model for characterizing variability in cryo-EM.

Area of Science:

• Structural biology
• Biophysics
• Computational biology

Background:

• Protein structural flexibility and dynamic interactions are crucial for function.
• Cryogenic electron microscopy (cryo-EM) visualizes macromolecules in various states.
• Current computational methods struggle with continuous conformational changes and unsupervised classification.

Purpose of the Study:

• To develop an automated machine learning algorithm for characterizing protein conformational landscapes.
• To provide a more intuitive and flexible representation of structural heterogeneity compared to existing manifold methods.

Main Methods:

• Developed e2gmm, a machine learning algorithm utilizing a deep neural network.
• Employs a three-dimensional Gaussian mixture model mapped onto 2D particle images with known orientations.
• Applied to simulated data and three biological systems.

Main Results:

• e2gmm automatically resolves structural heterogeneity in protein complexes.
• It maps particles onto a low-dimensional latent space representing conformational and compositional changes.
• Demonstrated effectiveness on diverse biological systems at various scales.

Conclusions:

• e2gmm offers an advanced, unsupervised approach to analyzing protein conformational dynamics from cryo-EM data.
• The algorithm provides a flexible and intuitive representation of structural landscapes.
• This method enhances the characterization of protein heterogeneity and conformational variability.