Omokage Search and Gmfit: Shape Similarity Search and Superposition among Models and Maps

Single particle CryoEM typically produces 3D density maps of the macromolecular complex with low resolution. For understanding these maps, it is essential to fit atomic models of individual subunits into the maps of the complex. We developed a program gmfit employing Gaussian mixture model (GMM) to rapidly superimpose subunits into the map (Kawabata, 2008, Biophys. J. 95, 4643-4658). The web server (“pairwise gmfit”) for pairwise superposition of atomic models and density maps is now available. We also developed the web server for Omokage search, which is a global shape similarity search in both the PDB and EMDB (Suzuki et al., Bioinformatics, submitted). The Omokage search is rapidly performed using 1D distance rank profiles, found similar models or maps are superimposed onto the query model or map using the pairwise gmfit program.Fitting multiple subunits into the map is more difficult than fitting one subunit. We try to include several additional experimental restraints to obtain more realistic configurations for low-resolution maps. (1) Overlaps with density map and (2) repulsions are two basic restraints. In addition, we introduce additional restraints: (3) symmetries of the subunits and (4) proximities between subunits. We introduce three methods to generate subunit initial configurations: symmetric operation, segmentation fitting, and distance geometry. The segmentation fitting method is repeating the “segmentation” of the map for each subunit, and the “fitting” of each subunit to the corresponding segmented region. The distance geometry method is employed for generating subunit configurations satisfying the given proximities. The subunit configurations are refined using the gradient-based optimization with the restraint energy. These methods have been tested for several simulated maps and real maps registered in EMDB. The source code of gmfit is freely available.