Challenges for Intelligent Image Processing in Cryo-Electron Microscopy

Christoph Best

Dept. of Molecular Structural Biology, Max Planck Institut fuer Biochemie, Martinsried, Germany

Nov 9, 2006

Cryo-electron microscopy enable the imaging of macromolecular complexes and cellular structures in a near-natural state at molecular resolution. Recent developments in preparation, instrumentation, and automation carry the promise of imaging molecular structures at sub-nanometer resolutions in their native environment, as well as creating molecular maps of the macromolecular complexes in the living cell. These advancements pose unique new informatics problems in image processing. In particular, methods from machine learning and probabilistic modeling will play a large role in classifying images, combining them into three-dimensional structures, and extracting information from them. I will discuss several examples where modern informatics methods may improve electron microscopy such as model-free maximum-likelihood classification of projection images, particle picking through support vector machines, and 3D reconstruction from random projections using the Baum-Welch algorithm and Level Set methods.