“2D Class Averaging of Cryo-Electron Microscopy Images”

Zhizhen (Jane) Zhao, U of I – Urbana/Champaign

Cryo-electron microscopy (EM) single-particle reconstruction is the process by which 3D density maps are obtained from a set of low-dose cryo-EM images of individual macromolecules. The low signal-to-noise ratio (SNR) of individual particle images makes it difficult to evaluate the images in a particle stack and reconstruct ab initio models. 2D class averaging is a crucial step to increase the SNR of raw projection images by clustering and averaging noisy images from similar viewing directions. However, without prior knowledge of the shape of the macromolecule, it is challenging to accurately identify images with similar views at low SNR.

In this talk, I introduce our 2D class averaging procedure and the computational methods within the pipeline. We apply steerable principal component analysis to compress and denoise images. We construct rotational invariant features of 2D images and use a fast approximate nearest neighbor search algorithm to achieve efficient initial classification. The pairwise alignment of particle images are performed among the nearest neighbors. The computational complexity of the initial classification scales linearly with the number of images. At extremely low SNR, the initial nearest neighbor list contains outliers. Therefore, we further improve the results by taking into account the geometry of the data manifold and the linear transformations between nearest neighbor pairs. We show that our procedure succeeds at remarkably low SNRs.