Segment Map partitions volume data to create a surface model with one or more segmentation regions (specialized surface pieces) shown in different colors. Along with the Fit to Segments tool, Segment Map is part of the Segger package described in:

Quantitative analysis of cryo-EM density map segmentation by watershed and scale-space filtering, and fitting of structures by alignment to regions. Pintilie GD, Zhang J, Goddard TD, Chiu W, Gossard DC. J Struct Biol. 2010 Jun;170(3):427-38.

See also: Segger documentation at NCMI, Segger how-to at the Chimera website, Volume Viewer, Volume Filter, Color Zone, segment, mask, measure

Methodology

Segment Map performs watershed segmentation: a density map is partitioned so that each local maximum has its own region, and the boundaries between regions lie at the valleys between the local maxima. The goal is usually one region per protein or domain, but the initial segmentation often gives more numerous and smaller regions than desired, especially if the data are noisy. Segger provides various ways to combine regions after the initial segmentation:

• automatic grouping by scale-space filtering (smoothing the data), done by default
• interactive grouping by manual selection
• interactive grouping by connectivity
• interactive grouping using fitted molecules

Data Size and Preprocessing

For working with large density maps (> 500 Mbytes), a 64-bit version of Chimera and a machine with more than 4 Gbytes of memory are recommended.

It may be necessary to process a density map beforehand to allow segmentation to finish in a reasonable amount of time. Maps of size 256³ with grid spacing about 1/3 of the map resolution are manageable on current (2010) desktop computers, while larger maps may require smoothing and binning. Calculation of 100,000 watershed regions is reasonable, but calculation of a million is unusably slow. A map of size 256³ can easily have a million local maxima if it is noisy. Gaussian filtering and binning to reduce the number of local maxima can be performed with Volume Filter or the command vop. Gaussian filtering reduces resolution to about six times the Gaussian standard deviation. For visualization and analysis, it is desirable for a density map to be oversampled by three times in each dimension, so binning should aim for a grid spacing of approximately twice the Gaussian standard deviation.

If interactive rotation speed is poor after segmentation, see response time issues.

There are several ways to start Segment Map, a tool in the Volume Data category (including from the Volume Viewer Tools menu).

The map of interest should be chosen from the pulldown menu of open volume data sets next to Segment map. The threshold (contour level) of the map in Volume Viewer should be adjusted before segmentation, because the calculation will use only the voxels with density values above that level.

Clicking Segment partitions the chosen map by the watershed method plus smoothing and grouping as specified in the options. The segmentation result is shown as a surface model with one or more segmentation regions (specialized surface pieces) of different colors. The resulting number of regions is reported in the dialog and the Reply Log. The segmentation is given the same name as the corresponding map except with .seg appended.

The Current segmentation is indicated in the dialog and can be changed by choosing a different segmentation name from the pulldown menu. Designating a segmentation as the current segmentation shows it and hides any others. Grouping and ungrouping operations modify the current segmentation model rather than creating a new one. However, multiple segmentation models can be open at the same time, and may have been generated from different density maps or opened from previously saved segmentation files.

Grouping and ungrouping can be specified interactively by selection. With default Mouse preferences, Ctrl-click selects a region and Shift-Ctrl-click selects additional regions without erasing the pre-existing selection.

• If no segmentation regions are selected, clicking Ungroup backtracks to the previous stage of grouping of the current segmentation. By default, Segment automatically applies three rounds of smoothing and grouping; after that, the three intermediate stages could be accessed by clicking Ungroup three times. If regions are selected, Ungroup will backtrack to the previous stage of grouping for the selected region(s).
• If no segmentation regions are selected, Group traverses the stages in the opposite direction as Ungroup, or performs further grouping. If regions are selected, Group will combine the selected regions into a single group.

Clicking Options reveals additional settings (clicking the close button on the right hides them again):

• Smoothing steps [N] step size [M] voxels - how many rounds of automatic smoothing and grouping (default N=3) should be performed right after watershed segmentation when the Segment button is clicked, and how much smoothing to perform at each round (default M=1). If L is the minimum of the grid spacing along the three axes, the standard deviation of the Gaussian used for smoothing will be MxL in the first round and increase by MxL in each subsequent round. A single round consists of smoothing the data, then letting each local maximum point from the previous round move by steepest ascent to the location of a new local maximum. If two or more previous maxima converge on a new maximum, their regions from the previous round are coalesced into a single region. In general, the number of rounds N is the dominant parameter to adjust when fewer, larger regions are desired. A larger step size M may be beneficial for noisier maps. However, a smaller step size gives more gradual changes per round, thus more rounds to achieve a similar end result, and more intermediate stages that can be traversed with the Group and Ungroup buttons. After the automatic rounds have completed, clicking Group executes the minimum number of additional rounds needed to change the grouping further.
  
  
• Keep only regions having at least [minsize] voxels - throw out regions with fewer than minsize voxels (default 1, retaining all regions). This applies only to the initial segmentation, not after smoothing.
  
  
• Stop grouping if number of regions <= [minreg] - backtrack to the previous stage of automatic grouping if the latest round would give fewer than minreg regions (default 1). This is useful for generating a segmentation with no fewer regions than the known or suspected number of components in a complex.
  
  
• Display at most [maxreg] regions - to avoid degrading performance, limit the number of displayed regions to maxreg (default 2000). If a greater number of regions is generated, however, those which are not displayed will still be included in subsequent smoothing and grouping calculations. If response time is poor when a high number (approaching maxreg) of regions is shown, decreasing maxreg may help.
  
  
• Surface granularity [trianglesize] voxels - surfaces consist of triangles; smaller triangles give a smoother appearance, but more triangles take longer to render. The default trianglesize of 1 gives triangles comparable in size to the map grid spacing. A value of 2 will make the triangles twice as large (quadruple the area) and reduce their number by about a factor of 4. Increasing trianglesize is recommended if response time is poor.
  
  
• Group with mouse [button] allows grouping watershed regions connected at some density level, controlled interactively by mouse drag. Activating this option reassigns the indicated mouse button (which can be changed using the pulldown menu) from its normal function in the Mouse preferences to grouping by connectivity. If the density map has not been segmented already, the initial click on the map with the assigned button performs watershed segmentation using only the voxels with density values above the current threshold setting in Volume Viewer, without subsequent smoothing/grouping. If grouping has already been performed, some ungrouping may be needed before using the connectivity method, as it will not group watershed regions together if they are already in different groups. Watershed regions continuous with the clicked density are grouped and shown in the same color. The level of density used to determine connectivity can be adjusted by mouse drag. Clicking a watershed region with the assigned button and dragging up (down) shows what regions would be continuous with the first at more (less) permissive density cutoffs. The coloring updates automatically: regions that would be connected are shown in the same color as the first, and unconnected but adjacent regions are shown in their own group colors, or if not grouped, in gray. The group being adjusted with the mouse will not extend into a pre-existing group. Releasing the mouse button groups the matching-color regions. Clicking with the assigned button on the background displays all of the ungrouped watershed regions in the current segmentation, and clicking again displays the grouped ones as well. Any previously created group can be adjusted by clicking one of its constituent watershed regions with the assigned button and dragging as described above.

Segmenting a map generates segmentation regions and automatically assigns them several attributes. These attributes can be inspected and new attributes assigned using the Region Attributes dialog, opened by choosing Regions... Attributes table from the Segment Map menu. See also: Define Attribute

In the Region Attributes dialog, each row is a region, and each column is an attribute that can be shown or hidden using the Columns menu. The table can be sorted by the values in any displayed column by clicking the column header. Clicking the header once sorts the entries in order of increasing value and places an up arrowhead (triangle) in the header. Clicking again sorts the entries in decreasing order and places a down arrowhead (inverted triangle) in the header.

One or more regions (rows) in the table can be chosen with the left mouse button. A contiguous block of rows can be chosen by dragging, or by clicking on the first (or last) line and then Shift-clicking on the last (or first). Ctrl-click toggles the status of a single line. Choosing a region in the table selects it, and selecting a region in the graphics window will center and highlight the corresponding row in the table.

The automatically created attributes are read-only:

• region - region ID number
• grid points - number of density map grid points within the region
• grouped - number of regions grouped to form the region (next level in hierarchy, not necessarily the number of watershed regions)
• has surface - 1 if a surface has been made (not necessarily shown) for the region
• contacts - number of contacting regions
• edge distance - number of voxels to the nearest edge of the density map
• bounds (hidden by default) - grid indices bounding the region: i_min, j_min, k_min, i_max, j_max, k_max

• Set attribute [attr_name] to value [attr_value] or [snapshot]

The table can filtered to show only the rows meeting some attribute criterion:

• Filter list [criterion]

grid_points > 1500
note
contacts == 0 and note and "good" in note

Clicking Update refreshes the Region Attributes dialog to reflect any changes in the current segmentation or its groupings. Close dismisses the dialog, and Help shows this manual page in a browser window.

The contents (excluding images) of the rows and columns currently shown in the filtered table can be saved in a comma-separated format using File... Export in the Region Attributes dialog menu. All of the attributes (including images) can be saved along with other segmentation information using File... Save Segmentation in the Segment Map menu.

Segmentation regions can be selected or colored based on their attribute values using the Render/Select by Attribute tool.

Menu entries apply only to the current segmentation unless stated otherwise.

• Open segmentation... open a previously saved Segger segmentation file
• Save segmentation - save the segmentation to a Segger file, name/location previously specified
• Save segmentation as... save the segmentation to a Segger file, specify name/location
• Save selected regions to .mrc file... save density map masked by the selected segmentation regions to an MRC file (map dimensions set to the minimal box containing the regions)
• Save all regions to .mrc file... save density map masked by all segmentation regions collectively as an MRC file (map dimensions set to the minimal box containing the regions)
• Save each region to .mrc file... save density map masked by each segmentation region as a separate MRC file (map dimensions set to the minimal box containing the corresponding region)
• Close segmentation - close the current segmentation model (a surface model)
• Close all segmentations except displayed - close all hidden segmentation models
• Close all segmentations - close all segmentation models

• Show all - show all segmentation regions (surface pieces) at the current level of grouping
• Show only selected - show the selected segmentation regions, hide all others
• Show adjacent - show regions in contact with the selected regions
• Show grouping - show the watershed regions within groups rather than the larger group surfaces; affects the selected regions, or if none are selected, all regions; does not change the groupings, only how they are shown
• Unshow grouping - show the larger group surfaces rather than the watershed regions; affects the selected regions, or if none are selected, all regions
• Hide - undisplay the selected regions, or if none are selected, all regions
• Make transparent - make 55% transparent the selected regions, or if none are selected, all regions
• Make opaque - make opaque (0% transparent) the selected regions, or if none are selected, all regions
• Color density map - color the density map display to match all regions (subsequently changing the threshold will erase the coloring for volume displays in the surface or mesh but not solid style)
• Select groups - select all regions composed of grouped watershed regions
• Select boundary regions - select regions that contain at least one grid point within three voxels of the edge of the density map
• Invert selection - deselect the selected regions and vice versa
• Group selected - group the selected regions
• Ungroup selected - ungroup the selected regions
• Smooth and group - execute the minimum number of rounds of smoothing and grouping needed to produce further grouping (equivalent to clicking the Group button with nothing selected)
• Delete selected - delete the selected regions (cannot be undone)
• Enclosed volume - report the approximate total volume enclosed by the selected regions (number of grid points and ų value obtained by multiplying the number of grid points by the voxel volume); more precise measurements can be obtained with Measure and Color Blobs
• Mask map with selected - mask the density map with the selected regions to create a new map (copy the density values inside the regions, set values elsewhere to zero, but maintain the original map dimensions)
• Show axes for selected - show principal axes of inertia as arrows for each selected region
• Hide all axes - remove all principal axes arrows
• Attributes table... show table of region attributes

High number of surface triangles. Surfaces are drawn as many small triangles. The number of surface triangles can be reduced by increasing the granularity (triangle size). Doubling the value will reduce the number of triangles by approximately a factor of 4 and may improve rendering speed by a similar factor. Also, the display can be limited to a specified maximum number of regions.

Selected surfaces. Chimera shows selections with a green outline (assuming default Background preferences). Selection outlines can increase the time to draw a single frame as much as fivefold. If rotation is slow, avoid having objects selected when the selection is not needed.

Mac graphics memory. Interactive performance becomes very poor on Mac OS 10.6 machines (probably earlier Mac OS versions too) when more than ~5000 segmentation regions are shown. As mentioned above, the display can be limited to a specified maximum number of regions. The poor performance appears to be a system graphics memory management problem and affects both ATI and Nvidia graphics cards. Tests on a Linux system with ATI Radeon HD 4850 graphics easily handled 50,000 region surfaces, while the same graphics on a Mac became quite slow with just 5000 surfaces. It does not seem to depend on the size of each surface. Performance can be improved somewhat, allowing manipulation of about twice as many surfaces, by disabling use of OpenGL vertex buffer objects (VBO) in Chimera. This can be done using Debug Graphics Driver. Disabling VBO makes Chimera use an older technology, OpenGL display lists, which probably get converted to VBO internally. Apple is unlikely to fix this problem in the near future.