A generalized framework for the image formation model and disentanglement of CTF parameters in cryo-electron microscopy

Abstract

In cryo-electron microscopy (cryo-EM), the microscope has a point spread function, whose Fourier transform, known as the Contrast Transfer Function (CTF), is determined by various parameters with a physical meaning (including height of the sample, magnetic lens aberrations, operating voltage, amplitude contrast, etc.). Some of the parameters are under experimental control, but they can also contain some experimental uncertainty and fluctuate over the course of data collection. While it is routine to use point estimates for the CTF parameters in a simplified forward model of image formation, we characterize here the uncertainty of the CTF by (1) framing the forward model of image formation as a probabilistic graphical model, (2) deriving conditional probabilities that leverage analytical linking functions, and (3) compare these results with existing approaches that model noise in frequency space, such as the “B-factor envelope” and the “white/coloured noise model”.