Are length 0 segments allowed in NeuroML?

[pgleeson]

Question from @apdavison:

This model: https://neuroml-db.org/model_info?model_id=NMLCL000641 has adjacent segments with the same x, y, z, diameter coordinates (ids 1036 and 1037), which gives a segment of length zero.

Is this allowed in NeuroML (i.e. should the software handle it?) or does this make the file invalid?

Personally I can't see a reason why it should be invalid; there are a lot of SWC reconstructions which have adjacent points, which would get converted to zero length segments. This shouldn't be an issue for most visualisation applications, so no need for them to say that they can't visualise the cell if they see it's invalid.

The jnml -validate option could throw a warning when it sees these segments, but currently doesn't (it could be added here).

For individual simulators, they could have an issue with this, if they map each segment to a compartment (as Moose might), but for Neuron using cables/sections with multiple segments, it shouldn't matter as long as the section doesn't just have one segment.

So ideally it should be the application which loads the NML in (or the conversion/export code) which decides whether this is an issue.

[sanjayankur31]

I've added this to the new docs here:

NeuroML/Documentation@eae47ec

Permalink: https://docs.neuroml.org/Userdocs/FAQ.html#are-length-0-segments-allowed-in-neuroml

[halfflat]

I'm concerned about the interpretation of zero-length segments as spheres. Referring to the component types doc, the calculation of the Conditional Derived Variable surfaceArea seems to imply that any zero length segment might be regarded as representing a sphere.

This becomes yet more complicated, if the proximal point of the segment is taken from or interpolated from the parent segment, as this would make its sphericity contingent upon another element.

From the point of view of interpreting these morphologies, zero length segments are not problematic, but potentially (and especially, contingently) spherical segments are. Certainly I'd like to be able to discount the possibility! But if they do exist, I think the conditions under which a segment might denote a sphere need to be nailed down.

Following the NeuroMorpho.org soma guidelines, for example, we'd find it much more straightforward to be able to interpret NeuroML morphologies as describing only truncated cones and cylinders, and relegate the conversion from formats with spherical somata to external tools, but acknowledge that that may not be feasible given morphology descriptions extant in the wild.

[halfflat]

On a related note, should that surface area calculation in the non-zero length case take into account the proximal radius as well? Even in the zero-length case, if the proximal and distal radii differ, and the segment is not regarded as representing a sphere, the segment describes an annulus, which also has non-zero area.