Source code for cnmodel.morphology.hoc_reader

from neuron import h
import neuron
import collections
import numpy as np
import pyqtgraph as pg
import os
import re
import os.path

[docs]class HocReader(object): def __init__(self, hoc): """ Provides useful methods for reading hoc structures. Parameters ---------- hoc : :obj: `hoc` or str Either a hoc object that hs been already created, or a string that defines a hoc file name. """ self.file_loaded = False if isinstance(hoc, basestring): fullfile = os.path.join(os.getcwd(), hoc) if not os.path.isfile(fullfile): raise Exception("File not found: %s" % (fullfile)) success = neuron.h.load_file(1, fullfile) if success == 0: # indicates failure to read the file raise NameError("Found file, but NEURON load failed: %s" % (fullfile)) self.file_loaded = True self.h = h # save a copy of the hoc object itself. else: self.h = hoc # just use the passed argument self.file_loaded = True # geometry containers self.edges = None self.vertexes = None # all sections in the hoc {sec_name: hoc Section} self.sections = collections.OrderedDict() # {sec_name: index} indicates index into self.sections.values() where # Section can be found. self.sec_index = {} # {sec_name: [mechanism, ...]} self.mechanisms = collections.OrderedDict() # {sec_group_name: set(sec_name, ...)} self.sec_groups = {} # topology {section: (parent, [children,...])} self.topology = {} # populate self.sections, self.sec_index, and self.mechanisms self.read_section_info() # auto-generate section groups based on either hoc section lists, or # on section name prefixes. sec_lists = self.get_section_lists() sec_prefixes = self.get_section_prefixes() # Add groupings by section list if possible: if len(sec_lists) > 1: self.add_groups_by_section_list(sec_lists) # Otherwise, try section prefixes elif len(sec_prefixes) > 1: for group, sections in sec_prefixes.items(): self.add_section_group(group, sections) # generate topology self._generate_topology()
[docs] def get_section(self, sec_name): """ Get the section associated with the section name Parameters ---------- sec_name : str The name of the section object. Returns ------- The hoc Section object with the given name. """ try: return self.sections[sec_name] except KeyError: raise KeyError("No section named '%s'" % sec_name)
[docs] def get_section_prefixes(self): """ Go through all the sections and generate a dictionary mapping their name prefixes to the list of sections with that prefix. For example, with sections names axon[0], axon[1], ais[0], and soma[0], we would generate the following structure: {'axon': ['axon[0]', 'axon[1]'], 'ais': ['ais[0]'], 'soma': ['soma[0]']} """ prefixes = {} regex = re.compile('(?P<prefix>\w+)\[(\d*)\]') for sec_name in self.sections: g = regex.match(sec_name) if g is None: continue prefix = g.group('prefix') prefixes.setdefault(prefix, []).append(sec_name) return prefixes
[docs] def get_mechanisms(self, section): """ Get a set of all of the mechanisms inserted into a given section Parameters ---------- section : :obj: `NEURON section` The NEURON section object. Returns ------- A list of mechanism names """ return self.mechanisms[section]
[docs] def get_density(self, section, mechanism): """ Get density mechanism that may be found the section. mechanism is a list ['name', 'gbarname']. This is needed because some mechanisms do not adhere to any convention and may have a different kind of 'gbarname' than 'gbar<name>_mechname' returns the average of the conductance density, as that may range across different values in a section (e.g., can vary by segments) Parameters ---------- section : :obj: `NEURON section` The NEURON section object. mechanism : list mechanism is a list ['name', 'gbarname']. It is used to retrieve the mechanism density from HOC as `segment.name.gbarname`. Returns ------- Mean conductance of the selected mechanism in the section, averaged across all segments of the section. """ gmech = [] for seg in section: try: x = getattr(seg, mechanism[0]) mecbar = '%s_%s' % (mechanism[1], mechanism[0]) if mecbar in dir(x): gmech.append(getattr(x, mechanism[1])) else: print 'hoc_reader:get_density did not find the mechanism in dir x', dir(x) except NameError: return(0.) except: print 'hoc_reader:get_density failed to evaluate the mechanisms... ' raise # print gmech if len(gmech) == 0: gmech = 0. return np.mean(gmech)
[docs] def get_sec_info(self, section): """ Get the info of the given section modified from: neuronvisio Parameters ---------- section : :obj: `NEURON section` The NEURON section object. Returns ------- str containing the information, with html formatting. """ info = "<b>Section Name:</b> %s<br/>" %section.name() info += "<b>Length [um]:</b> %f<br/>" % section.L info += "<b>Diameter [um]:</b> %f<br/>" % section.diam info += "<b>Membrane Capacitance:</b> %f<br/>" % section.cm info += "<b>Axial Resistance :</b> %f<br/>" % section.Ra info += "<b>Number of Segments:</b> %f<br/>" % section.nseg mechs = [] for seg in section: for mech in seg: mechs.append(mech.name()) mechs = set(mechs) # Excluding the repeating ones mech_info = "<b>Mechanisms in the section</b><ul>" for mech_name in mechs: s = "<li> %s </li>" % mech_name mech_info += s mech_info += "</ul>" info += mech_info return info
[docs] def read_section_info(self): """ Read all the information about the sections in the current hoc file Stores the result in the mechanisms class variable. """ # Collect list of all sections and their mechanism names. self.sections = collections.OrderedDict() self.mechanisms = collections.OrderedDict() for i, sec in enumerate(self.h.allsec()): self.sections[sec.name()] = sec self.sec_index[sec.name()] = i mechs = set() for seg in sec: for mech in seg: mechs.add(mech.name()) self.mechanisms[sec.name()] = mechs
[docs] def hoc_namespace(self): """ Get a dict of the HOC namespace {'variable_name': hoc_object}. NOTE: this method requires NEURON >= 7.3 """ names = {} for hvar in dir(self.h): # look through the whole list, no other way try: # some variables can't be pointed to... if hvar in ['nseg', 'diam_changed', 'nrn_shape_changed_', 'secondorder', 'stoprun']: continue u = getattr(self.h, hvar) names[hvar] = u except: continue return names
[docs] def find_hoc_hname(self, regex): """ Find hoc names matching a pattern Parameters ---------- regex : str Regular expression (Python Re module) to search for. Returns ------- list The names of HOC objects whose *hname* matches regex. """ objs = [] ns = self.hoc_namespace() for n, v in ns.items(): try: hname = v.hname() if re.match(regex, hname): objs.append(n) except: continue return objs
[docs] def add_section_group(self, name, sections, overwrite=False): """ Declare a grouping of sections (or section names). Sections may be grouped by any arbitrary criteria (cell, anatomical type, etc). Parameters ---------- name : str name of the section group sections: list section names or hoc Section objects. """ if name in self.sec_groups and not overwrite: raise Exception("Group name %s is already used (use overwrite=True)." % name) group = set() for sec in sections: if not isinstance(sec, basestring): sec = sec.name() group.add(sec) self.sec_groups[name] = group
[docs] def get_section_group(self, name): """ Get a section group by name Parameters ---------- name : str name of the group (dendrite, for example) Returns ------- The set of section names in the group *name*. """ return self.sec_groups[name]
[docs] def get_section_lists(self): """ Search through all of the hoc variables to find those that are "SectionLists" """ return self.find_hoc_hname(regex=r'SectionList\[')
#ns = self.hoc_namespace() #return [name for name in ns if ns[name].hname().startswith('SectionList[')]
[docs] def add_groups_by_section_list(self, names): """ Add a new section groups from the hoc variables indicated in *names*. Parameters ----------- names : list List containing variable names as strings. Each name must refer to a list of Sections in hoc. If a dict is supplied instead, then it maps {hoc_list_name: section_group_name}. Side effects (modifies) ----------------------- calls add_section_group """ # if a list is supplied, then the names of groups to create are # exactly the same as the names of hoc lists. if not isinstance(names, dict): names = {name:name for name in names} for hoc_name, group_name in names.items(): var = getattr(self.h, hoc_name) self.add_section_group(group_name, list(var))
[docs] def get_geometry(self): """ modified from:neuronvisio Generate structures that describe the geometry of the sections and their segments (all segments are returned) Returns ------- vertexes : record array containing {pos: (x,y,z), dia, sec_id} for each segment. edges : array of pairs indicating the indexes of connected vertexes. Side effects ------------ Modifies vertexes and edges. """ # return cached geometry if this method has already run. if self.vertexes is not None: return self.vertexes, self.edges self.h.define_shape() # map segments (lines) to the section that contains them self.segment_to_section = {} vertexes = [] connections = [] for secid, sec in enumerate(self.sections.values()): x_sec, y_sec, z_sec, d_sec = self.retrieve_coordinate(sec) for i,xi in enumerate(x_sec): vertexes.append(((x_sec[i], y_sec[i], z_sec[i]), d_sec[i], secid)) indx_geom_seg = len(vertexes) - 1 if len(vertexes) > 1 and i > 0: connections.append([indx_geom_seg, indx_geom_seg-1]) self.edges = np.array(connections) self.vertexes = np.empty(len(vertexes), dtype=[ ('pos', float, 3), ('dia', float), ('sec_index', int)]) self.vertexes[:] = vertexes return self.vertexes, self.edges
[docs] def retrieve_coordinate(self, section): """Retrieve the coordinates of a section avoiding duplicates Parameters ---------- section : :obj: `NEURON section` The NEURON section object. Returns ------- array arrays of x, y, z, d for all the segments of the specified section. """ section.push() x, y, z, d = [],[],[],[] tot_points = 0 connect_next = False for i in range(int(self.h.n3d())): present = False x_i = self.h.x3d(i) y_i = self.h.y3d(i) z_i = self.h.z3d(i) d_i = self.h.diam3d(i) # Avoiding duplicates in the sec if x_i in x: ind = len(x) - 1 - x[::-1].index(x_i) # Getting the index of last value if y_i == y[ind]: if z_i == z[ind]: present = True if not present: k =(x_i, y_i, z_i) x.append(x_i) y.append(y_i) z.append(z_i) d.append(d_i) self.h.pop_section() return (np.array(x),np.array(y),np.array(z),np.array(d))
def _generate_topology(self): for name, sec in self.sections.items(): sref = self.h.SectionRef(sec=sec) parent = sref.parent().sec.name() if sref.has_parent() else None if name not in self.topology: self.topology[name] = [None, []] self.topology[name][0] = parent if parent is not None: if parent not in self.topology: self.topology[parent] = [None, []] self.topology[parent][1].append(name)
[docs] def get_branch(self, root): """ Return all sections in a branch, starting with root. Parameters ---------- root : :obj: `NEURON section` The NEURON section object. """ branch = [root] childs = [root] while len(childs) > 0: new_childs = [] for ch in childs: new_childs.extend(self.topology[ch][1]) childs = new_childs branch.extend(childs) return branch
[docs] def translate_branch(self, root, dx): """ Move the branch beginning at *root* by *dx*. Parameters ---------- root : :obj: `NEURON section` The NEURON section object. dx : array Which must be an array of length 3 defining the translation. """ self.get_geometry() dx[np.newaxis, :] for name in self.get_branch(root): sid = self.sec_index[name] mask = self.vertexes['sec_index'] == sid self.vertexes['pos'][mask] += dx
[docs] def make_volume_data(self, resolution=1.0, max_size=500e6): """ Using the current state of vertexes, edges, generates a scalar field useful for building isosurface or volumetric renderings. Parameters ---------- resolution: float, default=1.0 microns width (um) of a single voxel in the scalar field. max_size: int maximum allowed scalar field size (bytes). Returns ------- * 3D scalar field indicating distance from nearest membrane, * 3D field indicating section IDs of nearest membrane, * QTransform that maps from 3D array indexes to original vertex coordinates. """ vertexes, lines = self.get_geometry() maxdia = vertexes['dia'].max() # maximum diameter (defines shape of kernel) kernel_size = int(maxdia/resolution) + 3 # width of kernel # read vertex data pos = vertexes['pos'] d = vertexes['dia'] sec_id = vertexes['sec_index'] # decide on dimensions of scalar field mx = pos.max(axis=0) mn = pos.min(axis=0) diff = mx - mn shape = tuple((diff / resolution + kernel_size).astype(int)) # prepare blank scalar field for drawing size = np.dtype(np.float32).itemsize * shape[0] * shape[1] * shape[2] if size > max_size: raise Exception("Scalar field would be larger than max_size (%dMB > %dMB), resolution is%f" % (size/1e6, max_size/1e6, resolution)) scfield = np.zeros(shape, dtype=np.float32) scfield[:] = -1000 # array for holding IDs of sections that contribute to each area idfield = np.empty(shape, dtype=int) idfield[:] = -1 # map vertex locations to voxels vox_pos = pos.copy() vox_pos -= mn.reshape((1,3)) vox_pos *= 1./resolution # Define kernel used to draw scalar field along dendrites def cone(i,j,k): # value decreases linearly with distance from center of kernel. w = kernel_size / 2 return w - ((i-w)**2 + (j-w)**2 + (k-w)**2)**0.5 kernel = resolution * np.fromfunction(cone, (kernel_size,)*3) kernel -= kernel.max() def array_intersection(arr1, arr2, pos): """ Return slices used to access the overlapping area between two arrays that are offset such that the origin of *arr2* is a *pos* relative to *arr1*. """ s1 = [0]*3 s2 = [0]*3 t1 = [0]*3 t2 = [0]*3 pos = map(int, pos) for axis in range(3): s1[axis] = max(0, -pos[axis]) s2[axis] = min(arr2.shape[axis], arr1.shape[axis]-pos[axis]) t1[axis] = max(0, pos[axis]) t2[axis] = min(arr1.shape[axis], pos[axis]+arr2.shape[axis]) slice1 = (slice(t1[0],t2[0]), slice(t1[1],t2[1]), slice(t1[2],t2[2])) slice2 = (slice(s1[0],s2[0]), slice(s1[1],s2[1]), slice(s1[2],s2[2])) return slice1, slice2 # Draw lines into volume using *kernel* as the brush vox_pos[:,0] = np.clip(vox_pos[:,0], 0, scfield.shape[0]-1) vox_pos[:,1] = np.clip(vox_pos[:,1], 0, scfield.shape[1]-1) vox_pos[:,2] = np.clip(vox_pos[:,2], 0, scfield.shape[2]-1) for c in range(lines.shape[0]): i = lines[c, 0] j = lines[c, 1] p1 = vox_pos[i].copy() p2 = vox_pos[j].copy() diff = p2-p1 axis = np.argmax(np.abs(diff)) dia = d[i] nvoxels = abs(int(diff[axis]))+1 for k in range(nvoxels): kern = kernel + (dia/2.0) sl1, sl2 = array_intersection(scfield, kern, p1) # find the overlapping area between the field and the kernel idfield[sl1] = np.where(scfield[sl1] > kern[sl2], idfield[sl1], sec_id[i]) scfield[sl1] = np.where(scfield[sl1] > kern[sl2], scfield[sl1], kern[sl2]) dia += (d[j]-d[i]) / nvoxels p1 += diff / nvoxels # return transform relating volume data to original vertex data transform = pg.Transform3D() w = resolution * kernel_size / 2 # offset introduced due to kernel transform.translate(*(mn-w)) transform.scale(resolution, resolution, resolution) transform.translate(1, 1, 1) return scfield, idfield, transform