[MRG] Make cell morphology plots reflect actual cell size

[ntolley]

Closes #433.

[ntolley]

Going through the code, the issue here is that section['sec_name'].end_pts is not automatically updated when length or other attributes are modified. For example, modifying by net.cell_types['L5_pyramidal'].sections['apical_trunk']._L = 5

In the code however, the length attribute is used by default when building the cell:

hnn-core/hnn_core/cell.py

Line 416 in bab8196

# be explicit about letting sec.L dominate over the 3d points used by

In my mind, the easiest solution is to somehow extract the real section.end_pts that Neuron calculates. Ideally without running a whole simulation though and building the whole network.

[rythorpe]

Wouldn't we then need to restrict plotting morphology until after NetworkBuilder is called?

[ntolley]

Wouldn't we then need to restrict plotting morphology until after NetworkBuilder is called?

I was thinking more in line with building the cell in Neuron, updating the values, and then deleting the object. Alternatively, we could make the interface more flexible and explicitly build the whole network without simulating, Then we could allow simulate_dipole to accept either a Network, or a NetworkBuilder object.

[jasmainak]

You could also copy the logic from here:

https://github.com/neuronsimulator/nrn/blob/f242cc8d4ddefb74f71f7f613a4b7b4043c2b641/src/nrnoc/treeset.cpp#L1233-L1253

Probably easier than rebuilding and destroying neuron objects ... but that trick might be useful elsewhere :)

[jasmainak]

Looks clean to me. @rythorpe merge button is yours when ready from @ntolley

[rythorpe]

I think there's still a bit more to do here, like call cell.change_sec_length() whenever a cell's section length attribute is modified. This could be done within Section.

[ntolley]

@rythorpe is right, there's a few things to take care of still. Good suggestion moving it to be a method of Section!

Right now the plots do update. However, all of the neighboring sections remain in place so the pieces are now disconnected. Cell.topology will specify the sections that need to be displaced. If I remember correctly, everything is referenced to the soma, so it should suffice to only move the "child_sec" of the sections being modified (and do so recursively).

[ntolley]

After working at this more, I honestly think building the cell is the best way to actually plot the morphology. Inspecting the geometry more closely, the points defined in Section.end_pts don't even match what's built in Neuron for the default cells.

With a built cell you can compare the two with

print(net.cell_types['L5_pyramidal'].sections['apical_trunk'].end_pts)
print(net.cell_types['L5_pyramidal']._nrn_sections['apical_trunk'].psection()['morphology']['pts3d'])

[[0, 0, 23], [0, 0, 83]]
[(0.0, 0.0, 39.0, 10.199999809265137), (0.0, 0.0, 141.0, 10.199999809265137)]

The length of the apical trunk is 102.0 so the Neuron coordinates are correct (attention that the z axis is in a different spot). I'll see if things start breaking when I create/delete cells in Neuron, but my sense is that this is a lot better than copying a bunch of C code from Neuron and translating it to python.

This also raises the question, why is there such a mismatch with the points defined here?

[ntolley]

Different approach, I've implemented Cell.update_end_pts() so that the object actually stores the real data used in Neuron. Since we're not starting any simulations with _PC I think that del is sufficient to clear the Neuron objects. I can confirm that the morphology plots now accurately reflect the actual cell size:

net = jones_2009_model()
#Default end_pts
net.cell_types['L5_pyramidal'].plot_morphology()

# Actual end_pts
net.cell_types['L5_pyramidal'].update_end_pts()
net.cell_types['L5_pyramidal'].plot_morphology()

# Modified end_pts
net.cell_types['L5_pyramidal'].sections['soma'].L = 500
net.cell_types['L5_pyramidal'].update_end_pts()
net.cell_types['L5_pyramidal'].plot_morphology()

[codecov-commenter]

Codecov Report

Merging #481 (ae49377) into master (c7460ba) will increase coverage by 0.10%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #481      +/-   ##
==========================================
+ Coverage   87.63%   87.74%   +0.10%     
==========================================
  Files          19       19              
  Lines        3792     3826      +34     
==========================================
+ Hits         3323     3357      +34     
  Misses        469      469              

Impacted Files	Coverage Δ
hnn_core/cell.py	97.21% <100.00%> (+0.43%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c7460ba...ae49377. Read the comment docs.

[jasmainak]

I'm not a big fan of the 2-step API ... can you make it so that everything is consistent as soon as you update the lengths? The user shouldn't have to remember to call another method for the lengths to be consistent.

[ntolley]

So I have made it so that all attribute modifications go through the common function Cell.modify_section(). Now every time that function is called, or a cell is created, Cell._update_end_pts() is called.

While before it was nice to be able to directly modify section attributes like Section.Ra and Section.cm, since we're making an API for length and diameter then we may as well have all attributes get changes in the same place.

[jasmainak]

agreed about that. Consistency in the API is important. Make sure to update whats_new when ready!

[jasmainak]

Thanks @ntolley ! 🥳