Outlier detection for single particle analysis in Electron Microscopy

Electron Microscopy (EM) of macromolecular structures using a single particle approach normally involves a two-dimensional (2D) classification step as a exploratory data analysis in which conformational changes, contaminants, or damaged particles may be identified. This step is nowadays even more important as automatic acquisition procedures are routinely employed and hundreds of thousands or millions of images can be acquired at the electron microscope. Automatic particle picking algorithms have a non-negligible false positive rate (wrongly selected particles), and many times they unadvertedly pass through the 2D classification, thus contaminating the dataset employed for 3D reconstruction. In this article we present an algorithm to reduce the number of these contaminating images, generally referred to as outliers.