Statistical modeling and removal of vesicles from cryo-EM images

Statistical modeling and removal of vesicles from cryo-EM images removes vesicle signals from cryo-electron microscopy micrographs to enable accurate alignment, classification, and high-resolution 3D reconstruction of membrane proteins within native lipid environments.

Key Features:

• Statistical Modeling: Employs higher-order singular value decomposition (HOSVD) using polar-coordinate symmetries to represent both round and non-round vesicle shapes.
• Non-linear Shape Alignment: Applies non-linear shape alignment to align vesicle morphologies with a reference model, preserving a compact statistical representation.
• Hierarchical Modeling: Summarizes multidimensional HOSVD coefficients through their principal components to capture vesicle asymmetries and enhance adaptability.
• Vesicle Signal Removal: Estimates the subspace occupied by vesicle structures and projects micrographs onto its orthogonal complement to remove interfering vesicle signals.
• Normalization and Model Selection: Integrates a robust normalization scheme and a model selection criterion to maintain compactness and generalizability across datasets.

Scientific Applications:

• Single-particle cryo-EM of membrane proteins: Removes vesicle background to improve particle alignment and classification for membrane protein datasets.
• High-resolution 3D reconstruction: Facilitates higher-resolution reconstructions of membrane protein assemblies, including complexes such as the Kv1.2 potassium-channel complex.

Methodology:

Constructs a 2D statistical vesicle model via HOSVD with polar-coordinate symmetries; applies non-linear shape alignment; summarizes multidimensional HOSVD coefficients by principal components (hierarchical modeling); projects micrographs onto the orthogonal complement of the vesicle subspace for signal removal; and employs a normalization scheme with model selection criterion.