def GetAllenCells(neuronal_model_ids):
from allensdk.api.queries.biophysical_api import BiophysicalApi
bp = BiophysicalApi()
bp.cache_stimulus = False # change to True to download the large stimulus NWB file
for neuronal_model_id in neuronal_model_ids:
os.system('mkdir models/' + str(neuronal_model_id))
bp.cache_data(neuronal_model_id,
working_directory=str(neuronal_model_id))
os.system('cd ' + str(neuronal_model_id) +
'; nrnivmodl ./modfiles; cd ../../')
def prepareModelRun(model_filename, ephys_filename, swc_filename, sweeps,
sweeps_by_type):
bp = BiophysicalApi()
bp.create_manifest(fit_path=os.path.basename(model_filename),
model_type='Biophysical - all active',
stimulus_filename=os.path.basename(ephys_filename),
swc_morphology_path=os.path.basename(swc_filename),
sweeps=sweeps)
bp.manifest['runs'][0]['sweeps_by_type'] = sweeps_by_type
manifest_file = os.path.join(os.path.dirname(model_filename),
'manifest.json')
with open(manifest_file, 'w') as manifest_json:
json.dump(bp.manifest, manifest_json, indent=2)
return manifest_file
def __init__(self, model_id, cache_stim=False):
self.model_id = model_id
self.bp = BiophysicalApi()
self.model_dir = os.path.join(os.getcwd(), 'model_' + str(model_id),
'')
self.bp.cache_stimulus = cache_stim
self.h = None
self.utils = None
self.description = None
self.section_output = {}
############################
self.reference_sweep = None
self.reference_output = None
self.j_history = []
self.theta_history = []
self.theta = None
self.theta_name = ''
self.fit_section = None
def model_props(cell_id):
cell_metadata = {}
# download existing model (if exists)
bp = BiophysicalApi()
try:
model_list = bp.get_neuronal_models(cell_id)
model_dict = {key: '' for key in template_model_dict.keys()}
for model_type, template_id in template_model_dict.items():
for model_meta in model_list:
if model_meta['neuronal_model_template_id'] == template_id:
model_dict[model_type] = model_meta['id']
except:
logger.debug('No biophysical model available')
api = allensdk.api.queries.rma_api.RmaApi() # Get the model metadata
for model_type, model_id in model_dict.items():
if model_id != '':
bp.cache_data(model_id, working_directory=model_dir[model_type])
model_metadata = api.model_query(
"NeuronalModelRun",
criteria="[neuronal_model_id$eq%s]" % model_id)
model_metadata_select = [
model_metadata[0]['rheobase_feature_average'],
model_metadata[0]['explained_variance_ratio']
]
model_path = glob.glob('%s/*fit*.json' % model_dir[model_type])[0]
for file_ in glob.glob('%s/*' % model_dir[model_type]):
if file_ != model_path:
try:
os.remove(file_)
except:
shutil.rmtree(file_, ignore_errors=True)
cell_metadata['model_path_{}'.format(
model_type)] = os.path.abspath(model_path)
cell_metadata['{}_id'.format(
model_dir[model_type])] = str(model_id)
for i, metadata_key in enumerate(opt_metric_dict[model_type]):
cell_metadata[metadata_key] = model_metadata_select[i]
return cell_metadata
def test_biophysical():
neuronal_model_id = 472451419 # get this from the web site
model_directory = "."
bp = BiophysicalApi("http://api.brain-map.org")
bp.cache_stimulus = False # don't want to download the large stimulus NWB file
bp.cache_data(neuronal_model_id, working_directory=model_directory)
os.system("nrnivmodl modfiles")
description = Config().load("manifest.json")
utils = Utils(description)
h = utils.h
manifest = description.manifest
morphology_path = manifest.get_path("MORPHOLOGY")
utils.generate_morphology(morphology_path.encode("ascii", "ignore"))
utils.load_cell_parameters()
stim = h.IClamp(h.soma[0](0.5))
stim.amp = 0.18
stim.delay = 1000.0
stim.dur = 1000.0
h.tstop = 3000.0
vec = utils.record_values()
h.finitialize()
h.run()
output_path = "output_voltage.dat"
junction_potential = description.data["fitting"][0]["junction_potential"]
mV = 1.0e-3
ms = 1.0e-3
output_data = (numpy.array(vec["v"]) - junction_potential) * mV
output_times = numpy.array(vec["t"]) * ms
DatUtilities.save_voltage(output_path, output_data, output_times)
assert numpy.count_nonzero(output_data) > 0
def test_biophysical():
neuronal_model_id = 472451419 # get this from the web site
model_directory = '.'
bp = BiophysicalApi('http://api.brain-map.org')
bp.cache_stimulus = False # don't want to download the large stimulus NWB file
bp.cache_data(neuronal_model_id, working_directory=model_directory)
os.system('nrnivmodl modfiles')
description = Config().load('manifest.json')
utils = Utils(description)
h = utils.h
manifest = description.manifest
morphology_path = manifest.get_path('MORPHOLOGY')
utils.generate_morphology(morphology_path.encode('ascii', 'ignore'))
utils.load_cell_parameters()
stim = h.IClamp(h.soma[0](0.5))
stim.amp = 0.18
stim.delay = 1000.0
stim.dur = 1000.0
h.tstop = 3000.0
vec = utils.record_values()
h.finitialize()
h.run()
output_path = 'output_voltage.dat'
junction_potential = description.data['fitting'][0]['junction_potential']
mV = 1.0e-3
ms = 1.0e-3
output_data = (numpy.array(vec['v']) - junction_potential) * mV
output_times = numpy.array(vec['t']) * ms
DatUtilities.save_voltage(output_path, output_data, output_times)
assert numpy.count_nonzero(output_data) > 0
def __init__(self, archive_dir=None):
self.bp = BiophysicalApi('http://api.brain-map.org')
self.bp.cache_stimulus = True # change to False to not download the large stimulus NWB file
self.cta = CellTypesApi()
self.rma = RmaApi()
self.neuronal_model_download_endpoint = 'http://celltypes.brain-map.org/neuronal_model/download/'
self.template_names = {}
self.nwb_list = []
if archive_dir == None:
archive_dir = '.'
self.archive_dir = archive_dir
from allensdk.api.queries.biophysical_api import \
BiophysicalApi
bp = BiophysicalApi()
bp.cache_stimulus = True # change to False to not download the large stimulus NWB file
neuronal_model_id = 472451419 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='neuronal_model')
from allensdk.api.queries.biophysical_api import \
BiophysicalApi
bp = BiophysicalApi()
bp.cache_stimulus = True # change to False to not download the large stimulus NWB file
neuronal_model_id = 472451419 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='neuronal_model')
pass
def getCells():
from allensdk.api.queries.biophysical_api import BiophysicalApi
bp = BiophysicalApi()
bp.cache_stimulus = True # change to False to not download the large stimulus NWB file
# # E2 cell (L2/3 Pyr) - https://senselab.med.yale.edu/modeldb/showmodel.cshtml?model=184161
neuronal_model_id = 472299294 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='E2full')
os.system('cd E2full; nrnivmodl ./modfiles; cd ..')
# E4 cell (L4 Pyr) - https://senselab.med.yale.edu/modeldb/showmodel.cshtml?model=184142
neuronal_model_id = 329321704 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='E4')
os.system('cd E4; nrnivmodl ./modfiles; cd ..')
os.system('cp %s %s' % ('E2/cell_template.hoc', 'E4/.'))
os.system('cp %s %s' % ('E2/cell_utils.py', 'E4/.'))
# E5 cell (L5 Pyr) - https://senselab.med.yale.edu/modeldb/showmodel.cshtml?model=184159
neuronal_model_id = 471087975 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='E5')
os.system('cd E5; nrnivmodl ./modfiles; cd ..')
os.system('cp %s %s' % ('E2/cell_template.hoc', 'E5/.'))
os.system('cp %s %s' % ('E2/cell_utils.py', 'E5/.'))
# IF cell (L5 Parv) - https://senselab.med.yale.edu/modeldb/showmodel.cshtml?model=184152
neuronal_model_id = 471085845 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='IF')
os.system('cd IF; nrnivmodl ./modfiles; cd ..')
os.system('cp %s %s' % ('E2/cell_template.hoc', 'IF/.'))
os.system('cp %s %s' % ('E2/cell_utils.py', 'IF/.'))
# IL cell (L5 Sst) - https://senselab.med.yale.edu/modeldb/showmodel.cshtml?model=184163
neuronal_model_id = 472299363 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='IL')
os.system('cd IL; nrnivmodl ./modfiles; cd ..')
os.system('cp %s %s' % ('E2/cell_template.hoc', 'IL/.'))
os.system('cp %s %s' % ('E2/cell_utils.py', 'IL/.'))
os.system('nrnivmodl ./E2/modfiles')
rma::criteria,[m__biophys_perisomatic$gt0],rma::criteria,[donor__species$il%27homo%20sapiens%27]'
tree = ET.fromstring(urllib.urlopen(url_spec).read())
neurons = []
# here we go through all the results and ask for the neuronal-model
for spec in tree.iter('specimen--id'):
url_model = "http://api.brain-map.org/api/v2/data/query.xml?criteria=model::NeuronalModel,\
rma::critera,[specimen_id$eq" + spec.text + "],neuronal_model_template[name$il'*Bio*']"
neurons.append(ET.fromstring(urllib.urlopen(url_model).read()).iter('id').next().text)
print("Found {0} neurons").format(len(neurons))
base_dir = os.getcwd()
working_directory = 'morphologies/ABI/' # where the model files will be downloaded
bp = BiophysicalApi('http://api.brain-map.org')
bp.cache_stimulus = False # set to True to download the large stimulus NWB file
for i, neuron in enumerate(neurons):
# os.mkdir(os.path.join(working_directory, neuron))
bp.cache_data(int(neuron), working_directory=os.path.join(working_directory, neuron))
# change directory
os.chdir(os.path.join(working_directory, neuron))
# compile and load NEURON NMODL files
os.system('nrnivmodl modfiles')
print("{} mechanism(s) compiled out of {}").format(i + 1, len(neurons))
os.chdir(base_dir)
# neuronal_model_id = 472451419 # get this from the web site as above
def download():
neuronal_model_ids = ['478047737', '472430904', '479234506', '478513451', '472352327', '478043748', '480633088', '478513443', '489931963', '478513224',
'478047816', '480632594', '472440759', '485694403', '478513459', '471081668', '477840124', '479234530', '473835796', '329322394',
'486557284', '479234670', '478513445', '472306616', '486509010', '478045347', '486509232', '480051220', '485611914', '480631402',
'472451419', '483108201', '496930338', '482525264', '479427516', '488461970', '477876583', '479427369', '472306544', '478810017',
'482657381', '478513187', '486511108', '479694721', '476630478', '486052412', '482655070', '472306460', '472442377', '478513396',
'482525598', '476637699', '487245869', '485510685', '478513398', '477494382', '482657528', '478513461', '485507735', '478513411',
'480629276', '473863510', '485591806', '488462820', '472447460', '480633809', '476637723', '478047588', '482524837', '477839613',
'478809991', '486556597', '480624051', '473465456', '472300877', '476679102', '480630211', '480633479', '480631048', '479427443',
'478045081', '477313251', '483156585', '329321704', '472299363', '480613522', '482934212', '485591776', '472434498', '472304676',
'488462783', '489932435', '473863578', '486052435', '486508758', '486556811', '473834758', '477880244', '478047618', '485602029',
'478513407', '485513184', '472421285', '477510985', '480631286', '491517388', '483066358', '479234441', '488763268', '473872986',
'488462965', '472455509', '489932597', '480624251', '480056382', '477876559', '478809612', '473863035', '480632564', '473862421',
'486556714', '485909152', '486508647', '473871773', '471086533', '482529696', '478045202', '489931686', '473465774', '471085845',
'480624414', '480631187', '480624126', '485476031', '480630857', '473862496', '471087975', '485909125', '472912177', '478513437',
'477838913', '478045226', '486558444', '478048947', '486909496', '477878284', '476630516', '484628276', '479694359', '485513255',
'484620556', '479694384', '483106906', '489932183', '485510712', '477510918', '472363762', '473862845', '486144663', '479695152',
'483109057', '478513415', '482583564', '488759006', '472912107', '485904766', '480630395', '480361288', '485720616', '477878554',
'476637796', '479694856', '478049069', '476637747', '472450023', '478513441', '472299294']
neuronal_model_ids = [472450023, 483108201, 486556811]
print("---- Downloading %i cell models..."%len(neuronal_model_ids))
for neuronal_model_id in neuronal_model_ids:
print("---- Downloading cell model: %s..."%neuronal_model_id)
try:
bp = BiophysicalApi('http://api.brain-map.org')
bp.cache_stimulus = False # change to False to not download the large stimulus NWB file
working_directory='%i'%neuronal_model_id
bp.cache_data(neuronal_model_id, working_directory=working_directory)
print("---- Saved model into %s, included NWB file: %s"%(working_directory,bp.cache_stimulus))
with open(working_directory+'/manifest.json', "r") as json_file:
manifest_info = json.load(json_file)
metadata={}
exp_id = int(manifest_info["biophys"][0]["model_file"][1][:9])
metadata['exp_id'] = exp_id
metadata['URL'] = 'http://celltypes.brain-map.org/mouse/experiment/electrophysiology/%s'%exp_id
for m in manifest_info['manifest']:
if m['key']=="output_path":
nwb_file = working_directory+'/'+m['spec']
if os.path.isfile(nwb_file):
print("---- Extracting metadate from NWB file: %s"%(nwb_file))
h5file=tables.open_file(nwb_file,mode='r')
metadata['AIBS:aibs_dendrite_type'] = str(h5file.root.general.aibs_dendrite_type.read())
metadata['AIBS:aibs_cre_line'] = str(h5file.root.general.aibs_cre_line.read())
metadata['AIBS:aibs_specimen_id'] = str(h5file.root.general.aibs_specimen_id.read())
metadata['AIBS:aibs_specimen_name'] = str(h5file.root.general.aibs_specimen_name.read())
metadata['AIBS:intracellular_ephys:Electrode 1:location'] = str(h5file.root.general.intracellular_ephys._v_children['Electrode 1'].location.read())
metadata['AIBS:session_id'] = str(h5file.root.general.session_id.read())
metadata['AIBS:subject:age'] = str(h5file.root.general.subject.age.read())
metadata['AIBS:subject:description'] = str(h5file.root.general.subject.description.read())
metadata['AIBS:subject:genotype'] = str(h5file.root.general.subject.genotype.read())
metadata['AIBS:subject:sex'] = str(h5file.root.general.subject.sex.read())
metadata['AIBS:subject:species'] = str(h5file.root.general.subject.species.read())
else:
print("---- Can't find NWB file: %s!"%(nwb_file))
print(' Metadata:')
pp.pprint(metadata)
with open(working_directory+'/metadata.json', 'w') as f:
json.dump(metadata, f, indent=4)
except IndexError:
print("Problem!")
def getCells():
from allensdk.api.queries.biophysical_api import BiophysicalApi
bp = BiophysicalApi()
bp.cache_stimulus = True # change to False to not download the large stimulus NWB file
# # E2 cell (L2/3 Pyr) - https://senselab.med.yale.edu/modeldb/showmodel.cshtml?model=184161
neuronal_model_id = 472299294 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='E2full')
os.system('cd E2full; nrnivmodl ./modfiles; cd ..')
# E4 cell (L4 Pyr) - https://senselab.med.yale.edu/modeldb/showmodel.cshtml?model=184142
neuronal_model_id = 329321704 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='E4')
os.system('cd E4; nrnivmodl ./modfiles; cd ..')
os.system('cp %s %s'%('E2/cell_template.hoc', 'E4/.'))
os.system('cp %s %s'%('E2/cell_utils.py', 'E4/.'))
# E5 cell (L5 Pyr) - https://senselab.med.yale.edu/modeldb/showmodel.cshtml?model=184159
neuronal_model_id = 471087975 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='E5')
os.system('cd E5; nrnivmodl ./modfiles; cd ..')
os.system('cp %s %s'%('E2/cell_template.hoc', 'E5/.'))
os.system('cp %s %s'%('E2/cell_utils.py', 'E5/.'))
# IF cell (L5 Parv) - https://senselab.med.yale.edu/modeldb/showmodel.cshtml?model=184152
neuronal_model_id = 471085845 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='IF')
os.system('cd IF; nrnivmodl ./modfiles; cd ..')
os.system('cp %s %s'%('E2/cell_template.hoc', 'IF/.'))
os.system('cp %s %s'%('E2/cell_utils.py', 'IF/.'))
# IL cell (L5 Sst) - https://senselab.med.yale.edu/modeldb/showmodel.cshtml?model=184163
neuronal_model_id = 472299363 # get this from the web site as above
bp.cache_data(neuronal_model_id, working_directory='IL')
os.system('cd IL; nrnivmodl ./modfiles; cd ..')
os.system('cp %s %s'%('E2/cell_template.hoc', 'IL/.'))
os.system('cp %s %s'%('E2/cell_utils.py', 'IL/.'))
os.system('nrnivmodl ./E2/modfiles')
from allensdk.api.queries.biophysical_api import BiophysicalApi
ids = [497232312, 491766131, 472451419]
bp = BiophysicalApi()
bp.cache_stimulus = True
for id in ids:
bp.cache_data(id, working_directory=f'model-{id}-new')
def download():
neuronal_model_ids = [
'478047737', '472430904', '479234506', '478513451', '472352327',
'478043748', '480633088', '478513443', '489931963', '478513224',
'478047816', '480632594', '472440759', '485694403', '478513459',
'471081668', '477840124', '479234530', '473835796', '329322394',
'486557284', '479234670', '478513445', '472306616', '486509010',
'478045347', '486509232', '480051220', '485611914', '480631402',
'472451419', '483108201', '496930338', '482525264', '479427516',
'488461970', '477876583', '479427369', '472306544', '478810017',
'482657381', '478513187', '486511108', '479694721', '476630478',
'486052412', '482655070', '472306460', '472442377', '478513396',
'482525598', '476637699', '487245869', '485510685', '478513398',
'477494382', '482657528', '478513461', '485507735', '478513411',
'480629276', '473863510', '485591806', '488462820', '472447460',
'480633809', '476637723', '478047588', '482524837', '477839613',
'478809991', '486556597', '480624051', '473465456', '472300877',
'476679102', '480630211', '480633479', '480631048', '479427443',
'478045081', '477313251', '483156585', '329321704', '472299363',
'480613522', '482934212', '485591776', '472434498', '472304676',
'488462783', '489932435', '473863578', '486052435', '486508758',
'486556811', '473834758', '477880244', '478047618', '485602029',
'478513407', '485513184', '472421285', '477510985', '480631286',
'491517388', '483066358', '479234441', '488763268', '473872986',
'488462965', '472455509', '489932597', '480624251', '480056382',
'477876559', '478809612', '473863035', '480632564', '473862421',
'486556714', '485909152', '486508647', '473871773', '471086533',
'482529696', '478045202', '489931686', '473465774', '471085845',
'480624414', '480631187', '480624126', '485476031', '480630857',
'473862496', '471087975', '485909125', '472912177', '478513437',
'477838913', '478045226', '486558444', '478048947', '486909496',
'477878284', '476630516', '484628276', '479694359', '485513255',
'484620556', '479694384', '483106906', '489932183', '485510712',
'477510918', '472363762', '473862845', '486144663', '479695152',
'483109057', '478513415', '482583564', '488759006', '472912107',
'485904766', '480630395', '480361288', '485720616', '477878554',
'476637796', '479694856', '478049069', '476637747', '472450023',
'478513441', '472299294'
]
neuronal_model_ids = [472450023, 483108201, 486556811]
print("---- Downloading %i cell models..." % len(neuronal_model_ids))
for neuronal_model_id in neuronal_model_ids:
print("---- Downloading cell model: %s..." % neuronal_model_id)
try:
bp = BiophysicalApi('http://api.brain-map.org')
bp.cache_stimulus = False # change to False to not download the large stimulus NWB file
working_directory = '%i' % neuronal_model_id
bp.cache_data(neuronal_model_id,
working_directory=working_directory)
print("---- Saved model into %s, included NWB file: %s" %
(working_directory, bp.cache_stimulus))
with open(working_directory + '/manifest.json', "r") as json_file:
manifest_info = json.load(json_file)
metadata = {}
exp_id = int(manifest_info["biophys"][0]["model_file"][1][:9])
metadata['exp_id'] = exp_id
metadata[
'URL'] = 'http://celltypes.brain-map.org/mouse/experiment/electrophysiology/%s' % exp_id
for m in manifest_info['manifest']:
if m['key'] == "output_path":
nwb_file = working_directory + '/' + m['spec']
if os.path.isfile(nwb_file):
print("---- Extracting metadate from NWB file: %s" %
(nwb_file))
h5file = tables.open_file(nwb_file, mode='r')
metadata['AIBS:aibs_dendrite_type'] = str(
h5file.root.general.aibs_dendrite_type.read())
metadata['AIBS:aibs_cre_line'] = str(
h5file.root.general.aibs_cre_line.read())
metadata['AIBS:aibs_specimen_id'] = str(
h5file.root.general.aibs_specimen_id.read())
metadata['AIBS:aibs_specimen_name'] = str(
h5file.root.general.aibs_specimen_name.read())
metadata[
'AIBS:intracellular_ephys:Electrode 1:location'] = str(
h5file.root.general.intracellular_ephys.
_v_children['Electrode 1'].location.read())
metadata['AIBS:session_id'] = str(
h5file.root.general.session_id.read())
metadata['AIBS:subject:age'] = str(
h5file.root.general.subject.age.read())
metadata['AIBS:subject:description'] = str(
h5file.root.general.subject.description.read())
metadata['AIBS:subject:genotype'] = str(
h5file.root.general.subject.genotype.read())
metadata['AIBS:subject:sex'] = str(
h5file.root.general.subject.sex.read())
metadata['AIBS:subject:species'] = str(
h5file.root.general.subject.species.read())
else:
print("---- Can't find NWB file: %s!" % (nwb_file))
print(' Metadata:')
pp.pprint(metadata)
with open(working_directory + '/metadata.json', 'w') as f:
json.dump(metadata, f, indent=4)
except IndexError:
print("Problem!")
def biophys_api():
endpoint = 'http://twarehouse-backup'
return BiophysicalApi(endpoint)
s = os.path.join(src, item)
d = os.path.join(dst, item)
if os.path.isdir(s):
shutil.copytree(s, d, symlinks, ignore)
else:
shutil.copy2(s, d)
#%% Pick a cell and a sweep to run from the available set of protocols
cell_id = 468193142 # get this from the web site: http://celltypes.brain-map.org
sweep_num = 46 # Select a Long Square sweep : 1s DC
#%% Download all-acive model
sdk_model_templates = {'all_active': 491455321, 'perisomatic': 329230710}
bp = BiophysicalApi()
bp.cache_stimulus = True
model_list = bp.get_neuronal_models(cell_id, model_type_ids=[sdk_model_templates['all_active']]) # Only get the all-active model for the cell
model_dict = model_list[0]
model_dir = 'all_active_models'
bp.cache_data(model_dict['id'], working_directory=model_dir)
new_model_file = 'fit_parameters_new.json'
shutil.copyfile(new_model_file, os.path.join(model_dir, new_model_file))
copytree('modfiles', os.path.join(model_dir, 'modfiles'))
#%% Running the legacy all-active models
os.chdir(model_dir)
subprocess.check_call(['nrnivmodl', 'modfiles/'])
manifest_file = 'manifest.json'
class Model(object):
'''
Class to manage:
- retrieving/setting up the data files from the AllenSDK
- running NEURON simulations
- fitting model parameters to experimental data
Parameters
----------
model_id: int
the id number that is associated with the biophysical model. Note that this is NOT the same as the cell id.
cache_stim: boolean
if set to true, this will cause the potentially-large NWB experimental data to be downloaded when the model is retrieved.
'''
def __init__(self, model_id, cache_stim=False):
self.model_id = model_id
self.bp = BiophysicalApi()
self.model_dir = os.path.join(os.getcwd(), 'model_' + str(model_id),
'')
self.bp.cache_stimulus = cache_stim
self.h = None
self.utils = None
self.description = None
self.section_output = {}
############################
self.reference_sweep = None
self.reference_output = None
self.j_history = []
self.theta_history = []
self.theta = None
self.theta_name = ''
self.fit_section = None
def __compile_modfiles(self):
'''
Helper function to compile modfiles in the ./modfiles/ folder
'''
#compilation on Windows
if os.name == 'nt' and not os.path.isfile('nrnmech.dll'):
# note that doing this programmatically is simpler on a *nix system, because of the nrnivmodl program.
# For compilation on Windows, the solution is usually to call
# %NEURONHOME%\bin\nrniv -nopython %NEURONHOME%\lib\hoc\mknrndll.hoc
# which brings up a GUI to select the modfiles folder, then produces a nrnmech.dll file.
# Here I'm directly calling the shell scripts that are called from the NEURON GUI. This
# works on my machines where 64-bit NEURON v7.4 was installed from binaries, but I would want to try it on
# other types of installations (eg: compiled from source) before I call it "stable."
shLoc = os.path.join(os.environ['NEURONHOME'], 'mingw', 'bin',
'sh.exe')
mknrnLoc = os.path.join(os.environ['NEURONHOME'], 'lib',
'mknrndll.sh')
nrnLoc = '/' + os.environ['NEURONHOME'].replace(':', '')
print "Attempting to compile modfiles..."
print[shLoc, mknrnLoc, nrnLoc]
os.chdir('modfiles')
subprocess.call([shLoc, mknrnLoc, nrnLoc])
os.chdir('..')
# copy the compiled file into the current directory to be used when we run NEURON
dllLoc = os.path.join('modfiles', 'nrnmech.dll')
if os.system("copy %s %s" % (dllLoc, '.')) == 0:
pass
else:
print "Could not compile modfiles using"
print shLoc, mknrnLoc, nrnLoc
print "Windows operating system detected. Please confirm that NEURON is installed appropriately for this usage."
quit()
# compilation on *nix systems
elif os.name != 'nt':
# compile the modfiles
subprocess.call(['nrnivmodl', 'modfiles'])
def init_model(self):
'''
Retrieves the model information for self.model_id, and initializes the modeling and visualization data.
'''
# Retrieve the model data
try:
self.bp.cache_data(self.model_id, working_directory=self.model_dir)
except:
print "Could not access biophysical data for model " + str(
self.model_id)
print "Please confirm model ID and try again."
quit()
# Compile the mod files for the current model
curr_dir = os.getcwd()
os.chdir(self.model_dir)
self.__compile_modfiles()
# set up the NEURON environment using the manifest file
description = Config().load('manifest.json')
# Specify model type (perisomatic/all-active)
currType = description.data['biophys'][0]['model_type']
# initialize the Hoc interpreter we'll use to affect the model
utils = create_utils(description, model_type=currType)
# configure the model for NEURON
morphologyLoc = description.manifest.get_path('MORPHOLOGY')
utils.generate_morphology(morphologyLoc.encode('ascii', 'ignore'))
utils.load_cell_parameters()
# store the hoc interpreter and cell information in the object instance
self.h = utils.h
self.utils = utils
self.description = description
# set up a monitor to log simulation time
self.section_output["t"] = self.h.Vector()
self.section_output["t"].record(self.h._ref_t)
# return to the previous directory
os.chdir(curr_dir)
def run_test_pulse(self, amp=0.18, delay=1000.0, dur=1000.0, tstop=3000.0):
'''
Performs a simple simulation with a square current injection pulse
Parameters
----------
amp: float
amplitude of stimulus (in nA)
delay: float
length of time to wait before applying stimulus (in ms)
dur: float
duration of the stimulus (in ms)
tstop: float
length of entire recording (in ms)
Returns
-------
vector
where vector['v'] is an array of floats indicating the response in mV
and vector['t'] is an array of floats indicating the time of each response point in ms.
'''
# generate response using a simple current-clamp stimulus
stim = self.h.IClamp(self.h.soma[0](0.5))
stim.amp = amp
stim.delay = delay
stim.dur = dur
self.h.tstop = tstop
# returns vector where vec['v'] = voltage, vec['t'] = time
self.model_output = self.utils.record_values()
self.h.finitialize()
self.h.run()
return self.model_output
def long_square(self, pulse_amp):
'''
Convenience function to run a model with stimulus of type "Long Square"
Parameters
----------
pulse_amp: float
amplitude of stimulus (in nA)
Returns
-------
vector
where vector['v'] is an array of floats indicating the response in mV
and vector['t'] is an array of floats indicating the time of each response point in ms.
'''
return self.run_test_pulse(amp=pulse_amp,
delay=100.0,
dur=1001.0,
tstop=1200.0)
def run_nwb_pulse(self, sweep_index):
'''
Performs a simulation using the NWB stimulus for the indicated sweep
Parameters
----------
sweep_index: int
indicates which sweep in the NWB file to use to generate the stimulus
Returns
-------
vector
where vector['v'] is an array of floats indicating the response in mV
and vector['t'] is an array of floats indicating the time of each response point in ms.
'''
if not self.bp.cache_stimulus:
print "Current model was not instantiated with NWB data cached. Please reload the current model and cache experimental stimulus data."
return
# set up the stimulus using the manifest data
stimulus_path = os.path.join(
self.model_dir,
self.description.manifest.get_path('stimulus_path'))
run_params = self.description.data['runs'][0]
sweeps = run_params['sweeps']
if not (sweep_index in sweeps):
print "Specified sweep index is not present in the current NWB dataset."
return
# run model with sweep stimulus
self.utils.setup_iclamp(stimulus_path, sweep=sweep_index)
self.model_output = self.utils.record_values()
self.h.finitialize()
self.h.run()
return self.model_output
def plot_output(self):
'''
Uses matplotlib to plot the data stored in self.model_output
'''
junction_potential = self.description.data['fitting'][0][
'junction_potential']
plt.plot(self.model_output['t'],
np.array(self.model_output['v']) - junction_potential)
plt.xlabel('time (ms)')
plt.ylabel('membrane potential (mV)')
plt.show()
def plot_fit(self):
'''
Uses matplotlib to plot the results of parameter fitting,
using the data stored in self.j_history and self.theta_history
'''
fig, theta_ax = plt.subplots()
cycles = range(len(self.theta_history))
theta_ax.plot(cycles, self.theta_history, 'b-')
theta_ax.set_ylabel('Theta', color='b')
theta_ax.set_xlabel('Cycle Number')
j_ax = theta_ax.twinx()
j_ax.plot(cycles, self.j_history, 'r-')
j_ax.set_ylabel('Cost', color='r')
plt.tight_layout()
plt.show()
def get_reconstruction(self):
'''
Checks for a morphology .swc file associated with the biophysical model.
If such a file is found, uses the AllenSDK swc helper functions to return a Morphology instance.
Returns
-------
Morphology
a morphology instance if a morphology file is found associated with the model. None otherwise.
'''
file_name = self.bp.ids['morphology'].values()[0]
if len(file_name) > 0:
file_name = os.path.join(os.getcwd(), self.model_dir, file_name)
return swc.read_swc(file_name)
else:
print "Could not find morphology file for model ", self.model_id, ", please check directory structure and try again."
return None
def monitor_section_voltage(self, sec_type, sec_index):
'''
Sets up a voltage monitor for the specified section.
Parameters
----------
sec_type: string
either 'soma', 'axon', or 'dend'
sec_index: int
indicates which section of sec_type to monitor the output from
'''
sec_name = sec_type + '[' + str(sec_index) + ']'
self.section_output[sec_name] = self.h.Vector()
if sec_type == 'dend':
self.section_output[sec_name].record(
self.h.dend[sec_index](0.5)._ref_v)
elif sec_type == 'axon':
self.section_output[sec_name].record(
self.h.axon[sec_index](0.5)._ref_v)
elif sec_type == 'soma':
self.section_output[sec_name].record(
self.h.soma[sec_index](0.5)._ref_v)
else:
print "Could not attach voltage monitor for section", sec_type, "[", sec_index, "]"
def set_fit_section(self, section_name, section_id=0):
'''
Sets the neuron compartment in which to perform parameter fitting.
Parameters
----------
section_name: string
either 'soma', 'axon', or 'dend'
section_id: int
index of the section of the type indicated by section_name
'''
if section_name == 'soma':
self.fit_section = self.h.soma[section_id]
elif section_name == 'axon':
self.fit_section = self.h.axon[section_id]
elif section_name == 'dend':
self.fit_section = self.h.dend[section_id]
else:
print "Could not set section for parameter fitting. Invalid section name."
self.fit_section = None
def set_parameter_to_fit(self, theta_name):
'''
Sets the parameter within the current section to be fit to the data.
Note that this function must be called after defining which section the parameter applies to.
Parameters
----------
theta_name: string
The name of the parameter (mechanism in NEURON terminology) to fit
'''
try:
self.theta_name = theta_name
if theta_name == 'gbar_NaV':
self.theta = self.fit_section.gbar_NaV
elif theta_name == 'gbar_Kv2like':
self.theta = self.fit_section.gbar_Kv2like
elif theta_name == 'gbar_Kv3_1':
self.theta = self.fit_section.gbar_Kv3_1
else:
print "Parameter fitting for mechanism", theta_name, "not yet implemented."
self.theta = None
self.theta_name = ''
except NameError:
print "Could not set parameter for fitting. Current section does not have a parameter/mechanism named", theta_name
self.theta = None
def change_fit_parameter(self, new_theta):
'''
Changes the value of the fit parameter.
Note that this must be called after set_parameter_to_fit().
Parameters
----------
new_theta: float
The new value of the parameter to assign.
'''
theta_name = self.theta_name
if theta_name == 'gbar_NaV':
self.fit_section.gbar_NaV = new_theta
self.theta = new_theta
elif theta_name == 'gbar_Kv2like':
self.fit_section.gbar_Kv2like = new_theta
self.theta = new_theta
elif theta_name == 'gbar_Kv3_1':
self.fit_section.gbar_Kv3_1 = new_theta
self.theta = new_theta
elif theta_name == None:
print "Attempting to assign a model parameter without specifying which parameter. Please call set_parameter_to_fit() first."
else:
print "Parameter fitting for mechanism", theta_name, "not yet implemented."
self.theta = None
self.theta_name = ''
def set_reference_sweep(self, ref_index):
'''
Checks to see if the given index refers to one of the sweeps associated with the model.
If it is, sets the sweep with this index as the reference sweep.
Parameters
----------
ref_sweep: int
the index of the reference sweep in the NWB file to compare the model to
'''
if not self.bp.cache_stimulus:
print "Current model was not instantiated with NWB data cached. Please reload the current model and cache experimental stimulus data."
return
# check for the existence of the requested sweep
stimulus_path = os.path.join(
self.model_dir,
self.description.manifest.get_path('stimulus_path'))
run_params = self.description.data['runs'][0]
sweeps = run_params['sweeps']
if not (ref_index in sweeps):
print "Specified sweep index is not present in the current NWB dataset."
self.reference_sweep = None
return
self.reference_sweep = ref_index
# Note that the allensdk.ephys.ephys_features.average_rate() function
# estimates average spike firing frequency by calculating (num spikes)/time.
# For sufficiently long recordings this should converge on the correct
# value, but for short pulses with a small number of spikes, you run
# into a granularity issue.
def average_rate_from_delays(self, t, spikes, start, end):
'''
Calculate average firing rate during interval between 'start' and 'end',
based on interspike interval durations.
Parameters
----------
t: numpy array of floats
times for each point in the recording (in seconds)
spikes: numpy array of ints
indices into the time/response arrays where a putative spike was detected
start: float
start of time window for spike detection (in seconds)
end: float
end of time window for spike detection (in seconds)
Returns
-------
float
average firing rate in spikes/second
'''
if start is None:
start = t[0]
if end is None:
end = t[-1]
n_spikes = len(spikes)
if n_spikes == 0:
return 0.0
prev_spike_t = start
sum_delays = 0.0
for spike_index in spikes:
curr_spike_t = t[spike_index]
sum_delays += curr_spike_t - prev_spike_t
prev_spike_t = curr_spike_t
return (float(n_spikes) / sum_delays)
def __J(self, model_val, exp_val, method='MSE'):
'''
Cost function to compare distance of model value from experimental value
Parameters
----------
model_val: numeric
the target value predicted by the model
exp_val: numeric
the target value observed in the experiment
method: string
defines the type of cost function to use. Currently only MSE is implemented
Returns
-------
float
the loss/cost associated with the difference between the model and the
experimental values
'''
if method == 'MSE':
return float(model_val - exp_val)**2
def set_up_objective(self, measure='spike frequency'):
'''
Prepares the model for parameter optimization by assigning the output measure to be used in the cost function.
Parameters
----------
measure: string
Name of the output measure to be used in optimization. Currently only 'spike frequency' is implemented.
'''
if (measure == 'spike frequency'):
# get the experimental data from the NWB file
data_set = NwbDataSet(
os.path.join(
self.model_dir,
self.description.manifest.get_path('stimulus_path')))
spike_times = data_set.get_spike_times(self.reference_sweep)
# calculate firing frequency for the NWB data
sum_intervals = 0.0
for i in range(len(spike_times) - 1):
sum_intervals += (spike_times[i + 1] - spike_times[i])
self.reference_output = len(spike_times) / sum_intervals
else:
print "Model fitting using the output measure", measure, "has not been implemented yet."
def objective_function(self,
theta,
measure='spike frequency',
stim_amplitude=0.21):
'''
Performs one simulation with the currently assigned theta set to the indicated value,
then returns the cost compared to the reference output based on the indicated cost measure.
Parameters
----------
theta: float
The value to assign to the current theta (stored at Model.theta_name)
measure: string
Name of the output measure to be used in optimization. Currently only 'spike frequency' is implemented.
stim_amplitude: float
The stimulation amplitude in nA to apply to simulate the reference sweep data.
Returns
-------
cost: float
A measure of difference between the reference sweep data and the current model
'''
if (measure == 'spike frequency'):
if self.theta_name == 'gbar_NaV':
self.fit_section.gbar_NaV = theta
elif self.theta_name == 'gbar_Kv2like':
self.fit_section.gbar_Kv2like = theta
elif self.theta_name == 'gbar_Kv3_1':
self.fit_section.gbar_Kv3_1 = theta
else:
print "Parameter fitting for mechanism", self.theta_name, "not yet implemented."
# Run the model
response = np.array(self.long_square(stim_amplitude)['v'])
times = np.array(self.model_output['t']) / 1000.0
startSec = 0.1
endSec = 1.1
spikes = detect_putative_spikes(response, times, startSec, endSec)
avg_rate = self.average_rate_from_delays(times, spikes, startSec,
endSec)
cost = self.__J(avg_rate, self.reference_output)
self.theta = theta
self.theta_history.append(theta)
self.j_history.append(cost)
return cost
def gradient_descent(self,
alpha=0.001,
epsilon=0.001,
threshold=0.001,
max_cycles=100):
'''
Attempts to minimize Model.objective_function() using an estimated gradient descent approach.
Iteratively performs simulations to optimize the current theta, which must be assigned before
this function is called using a call to set_parameter_to_fit().
Parameters
----------
alpha: float
The "learning rate" of the algorithm. Can be tuned empirically.
Lower values produce smoother, more stable optimization curves, but take longer to converge.
epsilon: float
Initial guess for amount to vary theta by. This will only be used in the initial estimate of
dJ/dTheta, and so should be set to a value that is small relative to theta.
threshold: float
The threshold for convergence. The function will return when the change in J for a given iteration falls below this value.
max_cycles: int
The maximum number of iterations to perform. The function will return regardless of values of J after this many iterations.
Returns
-------
cost: float
A measure of difference between the reference sweep data and the current model
'''
old_theta = self.theta
old_cost = self.objective_function(old_theta)
delta_cost = threshold + 1.0
cycle = 0
while (cycle < max_cycles) and (abs(delta_cost) > threshold):
new_theta = old_theta + epsilon
new_cost = self.objective_function(new_theta)
delta_cost = old_cost - new_cost
gradient = delta_cost / -epsilon
epsilon = -alpha * gradient
old_cost = new_cost
old_theta = new_theta
cycle += 1
print "theta", old_theta, "cost", old_cost, "gradient", gradient
