Cell Morphology

Simple morphologies can be written by hand using the native morphology format described below. For more complex structures the morphology can be read in either SWC format or in the XML format exported by Neuron. In particular, PSICS can read the structures provided by NeuroMorpho.org. The file must have the ".swc" extension and be placed in the same folder as the main xml file for the model. The ID of the morphology is taken to be the name of the file without the extension so, for example, to use the structure from a file called n128.swc the main model xml file would include the attribute specification morphology="n128".

The internal model used by PSICS differs somewhat from what is possible in SWC. This means that some features of the discretization and calculation process are not accessible for SWC morphologies. A partial solution is to use ICING to open the morphology and save it in the native format from which it can be edited by hand.

The main difference is that PSICS supports Genesis-style spherical compartments where branches are assumed to exit from the surface of the sphere with a radius set by the first daughter point (this is the function of the "minor" flag in the description below). By contrast points in swc files are always taken to be the ends of tapering segments so the soma of a cell, for example, is normally stored as several short fat segments with rapidly tapering segments where the dendrites emerge. This leads to possible mismatch between the physical extend of the soma and the computed area since segments that would physically intersect are treated as independent.

There other difference is in the labelling of points and sections of the structure. The different parts of a cell are often labelled in SWC files by setting a "type code" for each point. The code is often an integer, although it can also be a section if text as long as it does not contain any spaces. These codes are the only way in which points can be labelled, so it is not possible to also label a point as, for example, an intended recording location without overwriting its type code. Where the codes are used only for visualization, the effect is not very important, but it presents a problem if the codes are also used to assign channel densities to different regions of the cell.

In PSICS, the same approach is available via the "partof" property described below but the simplest way of specifying a region in a structure is to set labels only for its boundaries and then use structural relations to to define the regions. For example, the first point of the axon could be labelled "a0" and then the whole axon could be targeted as anything distal to "a0". Such expressions are not part of the morphology definition. Only the labels are present in the morphology file; thier significance comes with the CellProperties file that uses expressions to associate electrical properties with different regions of the cell. The benefit of working only with the boundaries is that they are easier to change by hand, and with a format like SWC that supports only a single label per point, it leaves most of the labels free for other uses such as specifying a recording site.

Thus although PSICS supports the "partof" attribute for points in a structure, this is mainly for backward compatibility and label based assignments are preferable. As well as being easier to create, they avoid a potential ambiguity in the boundaries of a region. For a set of points on a segment with a particular "partof" value specified, does the region just include the space between the first and the last or should it go half way back to the previous point? If it is the first one, what region does the segment between two points with different partof property belong to? Or in the second case, how do you specify where the soma ends and a dendrite begins? Putting a point at the boundary doesn't help because whatever property you set for it will extend half way to each neighbor. In practice, these issues are rarely important for realistic morphologies which are usually digitized to a fine enough resolution that there are several points to each compartment of the discretized structure. However, there is no reqirement in PSICS that a morphology specification should contain a lot of points (eg, a cell with a soma and uniform cables for an axon and a dendrite only needs three points) and it is for these idealized structures that the ambiguities of "partof" style specifications become problematic.

Native morphology format

The morphology of a cell is defined through the positions and radii of points on the structure. By default, the segment between neighboring points is assumed to taper smoothly where the radii differ at the two ends giving a carrotoid segment. This can be overridden with the "minor" flag which indicates that the distal segment starts from the surface of its parent and that its radius is fixed at the radius of the distal point. This typically applies to branches off the soma and to spines. It may also be appropriate for minor branches at branch points.

The Branch object is a convenience for defining minor branches without filling the parent segment with points. It has the same fields as the Point object, but is interpreted as a branch not directly from the parent point, but from somewhere along the segment proximal of the parent point such that the branch is perpendicular to the segment. This is convenient, for example, in modeling spines without modifying the specification of the segment to which they are attached. Note that for a branch you do not need to specify the position along the parent segment: this is computed internally so that the branch and its parent are at right angles. For branches at other angles it is necessary to insert a point in the parent segment where they join and then to use a normal point object for the child.

There are a two flags that can be used to make PSICS mimic the behavior of other systems. In the main PSICSRun file, setting squareCaps="true" causes the ends of segments to be ignored (no channels, no capacitance) rather than treated as hemispherical. In this case minor branches are treated as emerging from the center of the parent rather than the surface. The other flag is merge in the StructureDiscrtization element of the main PSICSRun file. If set to false, then there is at least one segment per point in the original structure file. The discretization process may still subdivide sections, but this can be avoided if desired by setting a large discretization parameter.