def normalize_all_nedms():
"""Iterate through all neuroelectro NeuronEphysDataMap objects and normalize for differences in units"""
nedms = m.NeuronEphysDataMap.objects.filter(neuron_concept_map__times_validated__gt = 0)
nedms = nedms.exclude(source__data_table__irrelevant_flag = True)
nedm_count = nedms.count()
for i,nedm in enumerate(nedms):
prog(i, nedm_count)
norm_dict = normalize_nedm_val(nedm)
# print norm_dict
# print nedm.val_norm
if nedm.val_norm is None:
# if no existing normalized value for nedm in DB
nedm.val_norm = norm_dict['value']
nedm.err_norm = norm_dict['error']
nedm.save()
elif np.isclose(nedm.val, nedm.val_norm):
# if there is a normalized value, but it's the same as the unnormalized value
# so it may need to be updated
norm_value = norm_dict['value']
if norm_value is None:
# can't normalize value but there's an existing one in the database
# so keep it as long as it's in an appropriate range
if check_data_val_range(nedm.val_norm, nedm.ephys_concept_map.ephys_prop) is False:
print 'deleting pre-normalized value of %s because out of appropriate range for %s ' % (nedm.val_norm, nedm.ephys_concept_map.ephys_prop )
nedm.val_norm = None
nedm.err_norm = None
nedm.save()
pass
elif np.isclose(norm_value, nedm.val_norm):
# new normalized value same as old normalized value, so do nothing
pass
# nedm.err_norm = norm_dict['error']
# nedm.save()
else:
# save nedm value
nedm.val_norm = norm_value
nedm.err_norm = norm_dict['error']
nedm.save()
# normalizing basically failed for some reason
else:
# there's a normalized value but it's different from what the algorithm suggests, so it's likely manually added
# if existing normalized value is out of range
if check_data_val_range(nedm.val_norm, nedm.ephys_concept_map.ephys_prop) is False:
print 'deleting pre-normalized value of %s because out of appropriate range for %s ' % (nedm.val_norm, nedm.ephys_concept_map.ephys_prop )
nedm.val_norm = None
nedm.err_norm = None
nedm.save()
else:
nedm.err_norm = norm_dict['error']
nedm.save()
# after all the checks above, do a final check for if algorithmically normalzied value is in correct range
annotate_misnormalized_nedm(nedm)
def test_check_data_val_range_out(self):
ephys_prop = m.EphysProp.objects.create(name = 'input resistance', min_range = .1, max_range = 10000)
data_val = -10
output_bool = check_data_val_range(data_val, ephys_prop)
expected_bool = False
self.assertEqual(output_bool, expected_bool)
def annotate_misnormalized_nedm(nedm):
''' if can't algorithmically normalize nedm value to something appropriate, and raw value is out of range,
leave a note in corresponding ecm in table'''
norm_dict = normalize_nedm_val(nedm)
if norm_dict['value'] is None and check_data_val_range(nedm.val, nedm.ephys_concept_map.ephys_prop) is False:
ecm = nedm.ephys_concept_map
normalizing_failed_note = 'Parsing failed to normalize ephys data'
if not ecm.note:
ecm.note = normalizing_failed_note
ecm.changed_by = m.get_robot_user()
ecm.save()
print 'adding failed normalizing note to %s with data table id %d' % (ecm.ephys_prop, ecm.source.data_table.pk)
def export_db_to_data_frame():
"""Returns a nicely formatted pandas data frame of the ephys data and metadata for each stored article"""
ncms = m.NeuronConceptMap.objects.all()#.order_by('-history__latest__history_date') # gets human-validated neuron mappings
ncms = ncms.exclude(Q(source__data_table__irrelevant_flag = True) | Q(source__data_table__needs_expert = True)) # exclude
ncm_count = ncms.count()
ephys_props = m.EphysProp.objects.all().order_by('-ephyspropsummary__num_neurons')
ephys_names = []
for e in ephys_props:
ephys_names.append(e.short_name)
ephys_names.append(e.short_name + '_err')
ephys_names.append(e.short_name + '_n')
ephys_names.append(e.short_name + '_sd')
ephys_names.append(e.short_name + '_note')
#ephys_names = [e.name for e in ephys_props]
#ncms = ncms.sort('-changed_on')
dict_list = []
for kk, ncm in enumerate(ncms):
prog(kk, ncm_count)
# TODO: need to check whether nedms under the same ncm have different experimental factor concept maps
# # check if any nedms have any experimental factors assoc with them
# efcms = ne_db.ExpFactConceptMap.objects.filter(neuronephysdatamap__in = nedms)
# for efcm in efcms:
# nedms = ne_db.NeuronEphysDataMap.objects.filter(neuron_concept_map = ncm, exp_fact_concept_map = ).distinct()
nedms = m.NeuronEphysDataMap.objects.filter(neuron_concept_map = ncm, expert_validated = True).distinct()
if nedms.count() == 0:
continue
sd_errors = identify_stdev(nedms)
temp_dict = dict()
temp_metadata_list = []
for nedm in nedms:
e = nedm.ephys_concept_map.ephys_prop
# check data integrity - value MUST be in appropriate range for property
data_val = nedm.val_norm
err_val = nedm.err_norm
n_val = nedm.n
note_val = nedm.ephys_concept_map.note
if check_data_val_range(data_val, e):
output_ephys_name = e.short_name
output_ephys_err_name = '%s_err' % output_ephys_name
output_ephys_sd_name = '%s_sd' % output_ephys_name
output_ephys_n_name = '%s_n' % output_ephys_name
output_ephys_note_name = '%s_note' % output_ephys_name
temp_dict[output_ephys_name] = data_val
temp_dict[output_ephys_err_name] = err_val
temp_dict[output_ephys_n_name] = n_val
temp_dict[output_ephys_note_name] = note_val
# do converting to standard dev from standard error if needed
if sd_errors:
temp_dict[output_ephys_sd_name] = err_val
else:
# need to calculate sd
if err_val and n_val:
sd_val = err_val * np.sqrt(n_val)
temp_dict[output_ephys_sd_name] = sd_val
#temp_metadata_list.append(nedm.get_metadata())
temp_dict['NeuronName'] = ncm.neuron.name
temp_dict['NeuronLongName'] = ncm.neuron_long_name
if ncm.neuron_long_name:
temp_dict['NeuronPrefName'] = ncm.neuron_long_name
else:
temp_dict['NeuronPrefName'] = ncm.neuron.name
article = ncm.get_article()
brain_reg_dict = get_neuron_region(ncm.neuron)
if brain_reg_dict:
temp_dict['BrainRegion'] = brain_reg_dict['region_name']
#article_metadata = normalize_metadata(article)
metadata_list = nedm.get_metadata()
out_dict = dict()
for metadata in metadata_list:
#print metadata.name
if not metadata.cont_value:
if metadata.name in out_dict:
out_dict[metadata.name] = '%s, %s' % (out_dict[metadata.name], metadata.value)
else:
out_dict[metadata.name] = metadata.value
elif metadata.cont_value and 'Solution' in metadata.name:
article = nedm.get_article()
amdm = m.ArticleMetaDataMap.objects.filter(article = article, metadata__name = metadata.name)[0]
ref_text = amdm.ref_text
out_dict[metadata.name] = ref_text.text.encode('utf8', "replace")
out_dict[metadata.name + '_conf'] = metadata.cont_value.mean
elif metadata.cont_value and 'AnimalAge' in metadata.name:
# return geometric mean of age ranges, not arithmetic mean
if metadata.cont_value.min_range and metadata.cont_value.max_range:
min_range = metadata.cont_value.min_range
max_range = metadata.cont_value.max_range
if min_range <= 0:
min_range = 1
geom_mean = np.sqrt(min_range * max_range)
out_dict[metadata.name] = geom_mean
else:
out_dict[metadata.name] = metadata.cont_value.mean
else:
out_dict[metadata.name] = metadata.cont_value.mean
# has article metadata been curated by a human?
afts = article.get_full_text_stat()
if afts and afts.metadata_human_assigned:
metadata_curated = True
metadata_curation_note = afts.metadata_curation_note
else:
metadata_curated = False
metadata_curation_note = None
if ncm.source.data_table:
data_table_note = ncm.source.data_table.note
else:
data_table_note = None
temp_dict2 = temp_dict.copy()
temp_dict2.update(out_dict)
temp_dict = temp_dict2
temp_dict['Title'] = article.title
temp_dict['Pmid'] = article.pmid
temp_dict['PubYear'] = article.pub_year
temp_dict['LastAuthor'] = unicode(get_article_last_author(article))
temp_dict['TableID'] = ncm.source.data_table_id
temp_dict['TableNote'] = data_table_note
temp_dict['ArticleID'] = article.pk
temp_dict['MetadataCurated'] = metadata_curated
temp_dict['MetadataNote'] = metadata_curation_note
#print temp_dict
dict_list.append(temp_dict)
base_names = ['Title', 'Pmid', 'PubYear', 'LastAuthor', 'ArticleID', 'TableID',
'NeuronName', 'NeuronLongName', 'NeuronPrefName', 'BrainRegion']
nom_vars = ['MetadataCurated', 'Species', 'Strain', 'ElectrodeType', 'PrepType', 'JxnPotential']
cont_vars = ['JxnOffset', 'RecTemp', 'AnimalAge', 'AnimalWeight', 'FlagSoln']
annot_notes = ['MetadataNote', 'TableNote']
for i in range(0, 1):
cont_vars.extend([ 'ExternalSolution', 'ExternalSolution_conf', 'external_%s_Mg' % i, 'external_%s_Ca' % i, 'external_%s_Na' % i, 'external_%s_Cl' % i, 'external_%s_K' % i, 'external_%s_pH' % i, 'InternalSolution', 'InternalSolution_conf', 'internal_%s_Mg' % i, 'internal_%s_Ca' % i, 'internal_%s_Na' % i, 'internal_%s_Cl' % i, 'internal_%s_K' % i, 'internal_%s_pH' % i])
#cont_var_headers.extend(['External_%s_Mg' % i, 'External_%s_Ca' % i, 'External_%s_Na' % i, 'External_%s_Cl' % i, 'External_%s_K' % i, 'External_%s_pH' % i, 'External_%s_text' % i, 'Internal_%s_Mg' % i, 'Internal_%s_Ca' % i, 'Internal_%s_Na' % i, 'Internal_%s_Cl' % i, 'Internal_%s_K' % i, 'Internal_%s_pH' % i, 'Internal_%s_text' % i])
col_names = base_names + nom_vars + cont_vars + annot_notes + ephys_names
# set up pandas data frame for export
df = pd.DataFrame(dict_list, columns = col_names)
# perform collapsing of rows about same neuron types but potentially across different tables
cleaned_df = df
# need to generate a random int for coercing NaN's to something - required for pandas grouping
rand_int = -abs(np.random.randint(20000))
cleaned_df.loc[:, 'Pmid':'FlagSoln'] = df.loc[:, 'Pmid':'FlagSoln'].fillna(rand_int)
grouping_fields = base_names + nom_vars + cont_vars
grouping_fields.remove('TableID')
cleaned_df.groupby(by = grouping_fields).mean()
cleaned_df.replace(to_replace = rand_int, value = np.nan, inplace=True)
cleaned_df.reset_index(inplace=True)
cleaned_df.sort_values(by = ['PubYear', 'Pmid', 'NeuronName'], ascending=[False, True, True], inplace=True)
cleaned_df.index.name = "Index"
# add in extra ephys data from columns based on known relationships, e.g., AP amp from AP peak and AP thr
cleaned_df = add_ephys_props_by_conversion(cleaned_df)
return cleaned_df
def export_db_to_data_frame():
"""Returns a nicely formatted pandas data frame of the ephys data and metadata for each stored article"""
ncms = (
m.NeuronConceptMap.objects.all()
) # .order_by('-history__latest__history_date') # gets human-validated neuron mappings
# ncms = ncms.exclude(Q(source__data_table__irrelevant_flag = True) | Q(source__data_table__needs_expert = True)) # exclude
ncms = ncms.exclude(Q(source__data_table__irrelevant_flag=True)) # exclude
ncm_count = ncms.count()
ephys_props = m.EphysProp.objects.all().order_by("-ephyspropsummary__num_neurons")
ephys_names = []
for e in ephys_props:
ephys_names.append(e.short_name)
ephys_names.append(e.short_name + "_raw")
ephys_names.append(e.short_name + "_err")
ephys_names.append(e.short_name + "_n")
ephys_names.append(e.short_name + "_sd")
ephys_names.append(e.short_name + "_note")
# ephys_names = [e.name for e in ephys_props]
# ncms = ncms.sort('-changed_on')
dict_list = []
for kk, ncm in enumerate(ncms):
prog(kk, ncm_count)
# TODO: need to check whether nedms under the same ncm have different experimental factor concept maps
# # check if any nedms have any experimental factors assoc with them
# efcms = ne_db.ExpFactConceptMap.objects.filter(neuronephysdatamap__in = nedms)
# for efcm in efcms:
# nedms = ne_db.NeuronEphysDataMap.objects.filter(neuron_concept_map = ncm, exp_fact_concept_map = ).distinct()
# only check whether ncms have been expertly validated, not the nedm itself
nedms = m.NeuronEphysDataMap.objects.filter(
neuron_concept_map=ncm, neuron_concept_map__expert_validated=True
).distinct()
if nedms.count() == 0:
continue
temp_dict = dict()
temp_metadata_list = []
for nedm in nedms:
e = nedm.ephys_concept_map.ephys_prop
# get error type for nedm by db lookup
error_type = nedm.get_error_type()
# check data integrity - value MUST be in appropriate range for property
data_val = nedm.val_norm
data_raw_val = nedm.val
err_val = nedm.err_norm
n_val = nedm.n
note_val = nedm.ephys_concept_map.note
output_ephys_name = e.short_name
output_ephys_raw_name = "%s_raw" % output_ephys_name
output_ephys_err_name = "%s_err" % output_ephys_name
output_ephys_sem_name = "%s_sem" % output_ephys_name
output_ephys_sd_name = "%s_sd" % output_ephys_name
output_ephys_n_name = "%s_n" % output_ephys_name
output_ephys_note_name = "%s_note" % output_ephys_name
# output raw vals and notes for all props
temp_dict[output_ephys_raw_name] = data_raw_val
temp_dict[output_ephys_note_name] = note_val
if check_data_val_range(data_val, e):
temp_dict[output_ephys_name] = data_val
temp_dict[output_ephys_err_name] = err_val
temp_dict[output_ephys_n_name] = n_val
# do converting to standard dev from standard error if needed
if error_type == "sd":
temp_dict[output_ephys_sd_name] = err_val
else:
# need to calculate sd
if err_val and n_val:
sd_val = err_val * np.sqrt(n_val)
temp_dict[output_ephys_sd_name] = sd_val
# temp_metadata_list.append(nedm.get_metadata())
temp_dict["NeuronName"] = ncm.neuron.name
temp_dict["NeuronLongName"] = ncm.neuron_long_name
if ncm.neuron_long_name:
temp_dict["NeuronPrefName"] = ncm.neuron_long_name
else:
temp_dict["NeuronPrefName"] = ncm.neuron.name
temp_dict["NeuroNERAnnots"] = ncm.get_neuroner()
article = ncm.get_article()
brain_reg_dict = get_neuron_region(ncm.neuron)
if brain_reg_dict:
temp_dict["BrainRegion"] = brain_reg_dict["region_name"]
# article_metadata = normalize_metadata(article)
metadata_list = nedm.get_metadata()
out_dict = dict()
for metadata in metadata_list:
# print metadata.name
if not metadata.cont_value:
if metadata.name in out_dict:
out_dict[metadata.name] = "%s, %s" % (out_dict[metadata.name], metadata.value)
else:
out_dict[metadata.name] = metadata.value
if metadata.name == "Strain":
out_dict["StrainNote"] = metadata.note
if metadata.name == "Species":
out_dict["SpeciesNote"] = metadata.note
elif metadata.cont_value and "Solution" in metadata.name:
article = nedm.get_article()
if metadata.ref_text:
ref_text = metadata.ref_text
else:
amdm = m.ArticleMetaDataMap.objects.filter(article=article, metadata__name=metadata.name)[0]
ref_text = amdm.ref_text
out_dict[metadata.name] = ref_text.text.encode("utf8", "replace")
out_dict[metadata.name + "_conf"] = metadata.cont_value.mean
elif metadata.cont_value and "AnimalAge" in metadata.name:
# return geometric mean of age ranges, not arithmetic mean
if metadata.cont_value.min_range and metadata.cont_value.max_range:
min_range = metadata.cont_value.min_range
max_range = metadata.cont_value.max_range
if min_range <= 0:
min_range = 1
geom_mean = np.sqrt(min_range * max_range)
out_dict[metadata.name] = geom_mean
else:
out_dict[metadata.name] = metadata.cont_value.mean
else:
out_dict[metadata.name] = metadata.cont_value.mean
# has article metadata been curated by a human?
afts = article.get_full_text_stat()
if afts and afts.metadata_human_assigned:
metadata_curated = True
metadata_curation_note = afts.metadata_curation_note
else:
metadata_curated = False
metadata_curation_note = None
if ncm.source.data_table:
data_table_note = ncm.source.data_table.note
else:
data_table_note = None
temp_dict2 = temp_dict.copy()
temp_dict2.update(out_dict)
temp_dict = temp_dict2
temp_dict["Title"] = article.title
temp_dict["Pmid"] = article.pmid
temp_dict["PubYear"] = article.pub_year
temp_dict["LastAuthor"] = unicode(get_article_last_author(article))
temp_dict["FirstAuthor"] = unicode(get_article_author(article, 0))
temp_dict["TableID"] = ncm.source.data_table_id
temp_dict["TableNote"] = data_table_note
temp_dict["ArticleID"] = article.pk
temp_dict["MetadataCurated"] = metadata_curated
temp_dict["MetadataNote"] = metadata_curation_note
# print temp_dict
dict_list.append(temp_dict)
base_names = [
"Title",
"Pmid",
"PubYear",
"FirstAuthor",
"LastAuthor",
"ArticleID",
"TableID",
"NeuronName",
"NeuronLongName",
"NeuronPrefName",
"NeuroNERAnnots",
"BrainRegion",
]
nom_vars = [
"MetadataCurated",
"Species",
"SpeciesNote",
"Strain",
"StrainNote",
"ElectrodeType",
"PrepType",
"JxnPotential",
]
cont_vars = ["JxnOffset", "RecTemp", "AnimalAge", "AnimalWeight", "FlagSoln"]
annot_notes = ["MetadataNote", "TableNote"]
grouping_fields = base_names + nom_vars + cont_vars
for i in range(0, 1):
cont_vars.extend(
[
"ExternalSolution",
"ExternalSolution_conf",
"external_%s_Mg" % i,
"external_%s_Ca" % i,
"external_%s_Na" % i,
"external_%s_Cl" % i,
"external_%s_K" % i,
"external_%s_pH" % i,
"external_%s_Cs" % i,
"external_%s_glucose" % i,
"external_%s_HEPES" % i,
"external_%s_EDTA" % i,
"external_%s_EGTA" % i,
"external_%s_BAPTA" % i,
"external_%s_ATP" % i,
"external_%s_GTP" % i,
"external_%s_CNQX" % i,
"external_%s_DNQX" % i,
"external_%s_NBQX" % i,
"external_%s_MK801" % i,
"external_%s_DAPV" % i,
"external_%s_CPP" % i,
"external_%s_kynur" % i,
"external_%s_BIC" % i,
"external_%s_picro" % i,
"external_%s_gabazine" % i,
"external_%s_CGP" % i,
"external_%s_strychnine" % i,
"InternalSolution",
"InternalSolution_conf",
"internal_%s_Mg" % i,
"internal_%s_Ca" % i,
"internal_%s_Na" % i,
"internal_%s_Cl" % i,
"internal_%s_K" % i,
"internal_%s_pH" % i,
"internal_%s_Cs" % i,
"internal_%s_glucose" % i,
"internal_%s_HEPES" % i,
"internal_%s_EDTA" % i,
"internal_%s_EGTA" % i,
"internal_%s_BAPTA" % i,
"internal_%s_ATP" % i,
"internal_%s_GTP" % i,
]
)
col_names = base_names + nom_vars + cont_vars + annot_notes + ephys_names
# not sure why but writing and reading data frame seems to fix a problem with ephys property pooling fxn
df = pd.DataFrame(dict_list, columns=col_names)
df.to_csv("temp.csv", sep="\t", encoding="utf-8")
df = pd.read_csv("temp.csv", sep="\t", index_col=0, header=0)
# perform collapsing of rows about same neuron types but potentially across different tables
# this should be optional if the goal is ephys recuration, not ephys reanalysis
grouping_fields.remove("TableID")
grouping_fields.remove("NeuroNERAnnots")
cleaned_df = pool_ephys_props_across_tables(df, grouping_fields)
# add in extra ephys data from columns based on known relationships, e.g., AP amp from AP peak and AP thr
cleaned_df = add_ephys_props_by_conversion(cleaned_df)
# returning 2 data frames, 1 with properties pooled and calculated based on algebra, 1 not
return cleaned_df, df
