# This file is part of Jaxley, a differentiable neuroscience simulator. Jaxley is
# licensed under the Apache License Version 2.0, see <https://www.apache.org/licenses/>
import networkx as nx
import numpy as np
from jaxley.io.graph import connect_graphs, from_graph
from jaxley.modules.base import to_graph
[docs]
def morph_delete(module_view) -> "Cell":
"""Deletes part of a morphology.
This function can only delete entire branches. It does not support deleting
compartments of a branch.
This function deletes all existing recordings, stimuli, and trainable parameters.
Args:
module_view: View of a `jx.Cell`. Defines the branches to be deleted.
Returns:
A cell in which specified branches are deleted.
Example usage
^^^^^^^^^^^^^
::
cell = jx.read_swc("path_to_swc_file.swc", ncomp=1)
cell = morph_delete(cell.axon)
::
cell = jx.Cell(branch, parents=[-1, 0, 0, 1, 1])
cell = morph_delete(cell.branch([3, 4]))
"""
# The `connect_graph` method cannot yet handle networks and branches.
assert module_view.base.__class__.__name__ == "Cell", (
f"You are trying to delete parts of a "
"`jx.{module_view.base.__class__.__name__}`. "
"Only `jx.Cell` is allowed in `morph_delete()`."
)
# Assert that there are no recordings, stimuli, and trainables.
assert len(module_view.base.recordings) == 0, (
f"Found {len(module_view.base.recordings)} recordings. This is not "
f"supported. Please run `cell.delete_recordings()`."
)
assert len(module_view.base.externals) == 0, (
f"Found {len(module_view.base.externals)} external states (stimuli or "
f"clamps). This is not supported. Please run `cell.delete_stimuli()` or "
f"`cell.delete_clamps()`."
)
assert len(module_view.base.trainable_params) == 0, (
f"Found {len(module_view.base.trainable_params)} trainable parameters. "
f"This is not supported. Please run `cell.delete_trainables()`."
)
# If the user did not run `compute_xyz` or `compute_compartment_centers`, we run
# it automatically.
if np.isnan(module_view.base.xyzr[0][0, 0]):
module_view.base.compute_xyz()
if "x" not in module_view.base.nodes.columns:
module_view.base.compute_compartment_centers()
comps_to_delete = module_view.nodes.index
comp_graph = to_graph(module_view.base)
nodes_to_keep = []
for node in comp_graph.nodes:
# We keep all `branchpoints`. The ones that have degree <2 will automatically
# be trimmed away `_remove_branch_points_at_tips`, which is run during
# `from_graph()`.
if (
comp_graph.nodes[node]["type"] == "branchpoint"
or comp_graph.nodes[node]["comp_index"] not in comps_to_delete
):
nodes_to_keep.append(node)
comp_graph = nx.subgraph(comp_graph, nodes_to_keep)
cell = from_graph(comp_graph)
return cell
[docs]
def morph_connect(module_view1, module_view2) -> "Cell":
"""Combine two morphologies into a single cell.
Both morphologies must have the same number of compartments per branch in all
branches.
This function deletes all existing recordings, stimuli, and trainable parameters.
Args:
module_view1: View of a ``jx.Cell()``. Must have been created with a
command ending on ``loc(0.0)`` or ``loc(1.0)``. For example, the following
are valid:
``cell.branch(0).loc(0.0)``, ``cell.branch(5).loc(0.0)``,
``cell.branch(5).loc(1.0)``.
But those are not valid:
``cell.branch(0).comp(0)`` (uses ``.comp``), ``cell.branch(5).loc(0.9)``
(does not use ``loc(0.0)`` or ``loc(1.0)``).
module_view2: The view of a ``jx.Cell()``. Must follow the same rules as
``module_view1``.
Returns:
A ``jx.Cell`` which is made up of both input cells.
Example usage
^^^^^^^^^^^^^
::
cell = jx.read_swc("path_to_swc_file.swc", ncomp=1)
stub = jx.Cell()
cell = morph_connect(cell.branch(0).loc(0.0), stub.branch(0).loc(0.0))
"""
# The `connect_graph` method cannot yet handle networks and branches.
for view in [module_view1, module_view2]:
assert view.base.__class__.__name__ == "Cell", (
f"You are trying to connect to a `jx.{view.base.__class__.__name__}`. "
"Only `jx.Cell` is allowed in `morph_connect()`."
)
# Assert that there are no recordings, stimuli, and trainables.
for view in [module_view1, module_view2]:
assert len(view.base.recordings) == 0, (
f"Found {len(view.base.recordings)} recordings. This is not "
f"supported. Please run `cell.delete_recordings()`."
)
assert len(view.base.externals) == 0, (
f"Found {len(view.base.externals)} external states (stimuli or "
f"clamps). This is not supported. Please run `cell.delete_stimuli()` or "
f"`cell.delete_clamps()`."
)
assert len(view.base.trainable_params) == 0, (
f"Found {len(view.base.trainable_params)} trainable parameters. "
f"This is not supported. Please run `cell.delete_trainables()`."
)
# If the user did not run `compute_xyz` or `compute_compartment_centers`, we run
# it automatically.
for view in [module_view1, module_view2]:
if np.isnan(view.base.xyzr[0][0, 0]):
view.base.compute_xyz()
if "x" not in view.base.nodes.columns:
view.base.compute_compartment_centers()
graph1 = to_graph(module_view1.base, channels=True)
graph2 = to_graph(module_view2.base, channels=True)
comps = []
for view in [module_view1, module_view2]:
global_comp = int(view.nodes.index.to_numpy()[0])
if len(view._comp_edges) == 0:
# Connect to the tip of a branch.
comps.append(global_comp)
else:
# Connect to a branchpoint.
# Branchpoint counters are stricly higher than compartment counters,
# so we can simply take the `max()` to get the branchpoint index.
comps.append(int(view._comp_edges["sink"].max()))
combined_graph = connect_graphs(graph1, graph2, comps[0], comps[1])
cell = from_graph(combined_graph)
return cell
