Versatility of Approximating Single-Particle Electron Microscopy Density Maps Using Pseudoatoms and Approximation-Accuracy Control

Table 1

Comparison of Vector Quantization, Gaussian Mixture Model, and our pseudoatomic model.

	Vector Quantization	Gaussian Mixture Model	Our pseudoatomic model
Grain	3D point (codebook vector)	3D Gaussian distribution function	3D radial basis function (isotropic Gaussian distribution function)
Grain geometry	Spherical	Ellipsoidal	Spherical
Algorithm	Self-organizing map (SOM)	Maximum likelihood method using the expectation maximization algorithm	Iterative adding and removing of pseudoatoms and gradient descent refinement
Goal of algorithm	Minimize the mean-square deviation of the codebook vectors from the corresponding 3D data	Find the model with the maximum likelihood function	Find the model with the minimum number of grains for the given error of density approximation
Number of grains	Fixed	Fixed	Adjustable
Grain weight	Adjustable	Adjustable	Adjustable
Grain position	Adjustable	Adjustable	Adjustable
Grain size	Adjustable	Adjustable	Fixed
Application of elastic network model	Easy	Difficult	Easy