Divisions of Computational Biology/ Gene Regulation & Expression
One of the biggest challenges in the analysis of cryo-EM images is the heterogeneity and flexibility of the molecules, which on the one hand severely limits the achievable resolution but on the other hand reports on conformational dynamics. A computational approach will be presented to reduce the conformational variance of a set of single-particle images with the goal of increasing the resolution. We reconstruct conformational variance of a molecule from the variance of the density and use this information for image classification. In a next step, the individual 3D density reconstructions from all classes are recombined into one single (higher resolution) reconstruction by a novel elastic map alignment procedure. The goal is to improve the resolution and at the same time to gain a complete picture of the conformational variance of a macromolecule. I will also present new computational tools to build and refine protein structures using intermediate-resolution density maps; in particular an automatic backbone tracing algorithm. To build accurate protein models at low-resolution requires to predict local structural details that are not resolved by the experiment. For this we have developed a restrained MD simulation approach and it is shown how the use of evolutionary information can significantly improve the refinement.