Use of nonlinear mapping in multivariate image analysis of molecule projections

A nonlinear mapping algorithm has been designed for the statistical analysis of sets of images in electron microscopy. The algorithm is appropriate in situations where it is not isolated features that determine the clustering or trends in the data set but a combination of several simultaneously changing features. The combination of the nonlinear mapping algorithm with correspondence analysis offers a way of ordering a set of images according to the variations of a single feature, as well as according to the combined variation of several independent features. The application of the combined algorithms ranges from image selection for averaging procedures to the sorting of images, that inherently form a tilt series. These sorted images then can be used for the three-dimensional reconstruction of macromolecules.