PMC

The methods taken in the development and evaluation of THESIS. Apps from systematic keyword search in Apple and Google stores (n = 3191).

Mark pixels as occupied and extending SWC structure in 3D. (A) Mark pixels in tiff stack as occupied. The pixels around two connected SWC points (red spheres), formed by two half cylinders and a trapezoidal prism, are marked as occupied. (B) The candidate point for extension is searched in the plane of an end SWC point (yellow circle). Red circles are the SWC points that are connected to the end point. The search is done along a shortest distance path running through the neurite (white line). This path is determined by building the intensity-weighted shortest distance profile along an arc (blue line) enclosed by two black lines. The valid SWC point closest to the end point is selected as the new SWC point (orange circle). (C) The profile of the shortest distances along the arc [blue line in (B)]. The green line is the smoothed version. The angle is measured relative to the line connecting the end point to its connected SWC point [yellow line in (B)]. The minimum position in the smoothed profile is selected as the starting point of the shortest path shown in (B).

The flowchart of our modeling procedure using MAINMAST. First, points with a high local density are identified in the query EM map using the mean shift algorithm. Identified local dense points are connected by minimum spanning tree (MST). Using tabu search, the initial MST is refined. Next, the amino acid sequence of the protein is mapped on the longest path in the tree by matching the volume of amino acids to the density of the local dense points (threading). Trees were generated with different combinations of parameters that control the local dense point identification step, the tree refinement step, and the sequence mapping result, which results in over a few thousand Cα models. The Cα models are then ranked with the density-volume matching (threading) score. The 500 top-scoring Cα models are selected, which are subject to full-atomic structure building with PULCHRA and structure refinement with Rosetta, MDFF, or xMDFF. Finally, the 500 full-atom models are ranked by the scoring function of the refinement method used and the top 10 scoring models were submitted to the Model Challenge.

Diagrams for the 3-D refinement procedure and the multi-path simulated annealing algorithm. The novel features are shown in the shaded boxes. (a) The refinement procedure includes three loops and one step to build an initial random model. The outer loop is used to iteratively refine a 3-D map. At the beginning of the outer loop, the data resolution range used in common line search can be chosen; and also the multipath SA algorithm can set to execute either global or local search. The inner loop is used to generate M candidate solutions which will be passed to the middle loop for consistency check. Each of the M solutions is produced by multi-path SA. The consistency measure identifies whether the raw particle image is good, in which case it is used in the corresponding iteration of 3-D refinement. The middle loop goes through each of the N raw particle images and then generates N′ (<N) particles with consistent solutions. The N′ particles are used for building a 3-D map using the make3d program of EMAN. (b) The multi-path SA algorithm runs in two modes: global alignment search and local alignment refinement. The loop goes through three annealing stages based on annealing temperature with a different path update scheme in the third stage for updating the current solution list. The current solution list is sorted by residual after each update. Paths are quasi-randomly picked, with some preference given to the worst residual paths in the first two stages. Once a path is picked, a trial solution is derived by randomly varying its centers and orientations. The trial residual is calculated based on the trial common lines. The acceptance of a trial solution is governed by the Metropolis criterion. An accepted trial solution replaces the corresponding solution in the current solution list in first two stages, or the worst solution in the list in the last stage. The best solution list always keeps track of the best solution found so far in each individual path. The final solution will be the best element in the best solution list.