Class Averaging of Cryo-Electron Microscopy Images
Friday, November 13, 2015
12:00 pm - 1:00 pm
Gross Hall 330
Zhizhen (Jane) Zhao, NYU-CAOS-Courant Institute
Abstract: Class averaging is a crucial initial step in cryo-electron microscopy single particle reconstruction, because the signal to noise ratio (SNR) of raw projection images is typically too low for ab inito modeling. Class Averaging amplifies the SNR by averaging noisy images of similar viewing directions. The averaged images form the input to ab initio reconstruction algorithms to determine the 3D electron density map of a macromolecule. Without prior knowledge of the particle, identifying images from similar viewing directions is challenging for large data sets at low SNR. Our class averaging procedure uses fast steerable PCA to compress and denoise images. We construct rotational invariant representation of 2D images and use randomized algorithms for dimensionality reduction and approximate nearest neighbor search to achieve efficient initial classification (near linear running time in the number of images). The initial classification is further improved by vector diffusion maps. We show that our procedure succeeds at remarkably low levels of signal-to-noise ratio. Lunch will be served.