Electron micrographs obtained from biological specimens generally have very low signal-to-noise ratios. To determine the structure of macromolecular assemblies and cellular organelles by electron microscopy, it is therefore necessary to employ image processing methods that must often be adapted for particular applications. To address such problems in our laboratory, we are developing and implementing a series of computational tools. Structures of filamentous macromolecules that exhibit tight and variable curvature can be obtained by first straightening them, computationally, prior to image averaging, classification, and three-dimensional reconstruction. We are developing a computer program that straightens filaments, having any curvature profile, by fitting non-uniform cubic-splines to their center lines. Two-dimensional crystalline specimens, in general, are liable to contain several different kinds of disorder that can be corrected computationally. Methods for two-dimensional lattice unwarping followed by averaging of unit cells are being applied to the membrane proteins prepared from urinary bladder epithelium of wild type and its knockout mutants to detect any potential rearrangement. In another image processing application, multivariate statistical methods are being applied to extract elemental maps from electron energy loss spectroscopic images in which spectra are collected at each pixel to give a three-dimensional data cube in x, y and energy loss.
