Automatic post-picking improves particle image detection from Cryo-EM micrographs

TL;DR

This paper introduces an automated post-picking classification method for cryo-EM micrograph images, significantly reducing manual effort and achieving human-like accuracy using ensemble classifiers and new image features.

Contribution

It presents a novel supervised post-processing step that classifies micrograph windows into particles or non-particles, improving efficiency over existing automated methods.

Findings

Reduces manual workload by orders of magnitude.

Achieves human-like classification performance with few training samples.

Uses new powerful image features and ensemble classifiers.

Abstract

Cryo-electron microscopy (cryo-EM) studies using single particle reconstruction is extensively used to reveal structural information of macromolecular complexes. Aiming at the highest achievable resolution, state of the art electron microscopes acquire thousands of high-quality images. Having collected these data, each single particle must be detected and windowed out. Several fully- or semi-automated approaches have been developed for the selection of particle images from digitized micrographs. However they still require laborious manual post processing, which will become the major bottleneck for next generation of electron microscopes. Instead of focusing on improvements in automated particle selection from micrographs, we propose a post-picking step for classifying small windowed images, which are output by common picking software. A supervised strategy for the classification of windowed micrograph images into particles and non-particles reduces the manual workload by orders of magnitude. The method builds on new powerful image features, and the proper training of an ensemble classifier. A few hundred training samples are enough to achieve a human-like classification performance.