In [1]:
%load_ext autoreload
%autoreload 2
In [2]:
import os
import expipe
import pathlib
import numpy as np
import spatial_maps.stats as stats
import septum_mec
import septum_mec.analysis.data_processing as dp
import septum_mec.analysis.registration
import head_direction.head as head
import spatial_maps as sp
import speed_cells.speed as spd
import re
import joblib
import multiprocessing
import shutil
import psutil
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib
from distutils.dir_util import copy_tree
from neo import SpikeTrain
import scipy
import statsmodels
import seaborn as sns
from tqdm.notebook import tqdm_notebook as tqdm
tqdm.pandas()
from spike_statistics.core import permutation_resampling_test
from spikewaveform.core import calculate_waveform_features_from_template, cluster_waveform_features
from septum_mec.analysis.plotting import violinplot, despine
In [3]:
#############################
zscore_str = "-no-zscore"
# zscore_str = ""
# stim_loc = 'mec'
stim_loc = 'ms'
#################################
In [4]:
%matplotlib inline
plt.rc('axes', titlesize=12)
plt.rcParams.update({
'font.size': 12,
'figure.figsize': (6, 4),
'figure.dpi': 150
})
output_path = pathlib.Path("output") / ("stimulus-lfp-response" + '-' + stim_loc + zscore_str)
(output_path / "statistics").mkdir(exist_ok=True, parents=True)
(output_path / "figures").mkdir(exist_ok=True, parents=True)
output_path.mkdir(exist_ok=True)
In [5]:
data_loader = dp.Data()
actions = data_loader.actions
project = data_loader.project
In [6]:
identify_neurons = actions['identify-neurons']
sessions = pd.read_csv(identify_neurons.data_path('sessions'))
In [7]:
lfp_action = actions['stimulus-lfp-response' + zscore_str]
lfp_results = pd.read_csv(lfp_action.data_path('results'))
In [8]:
lfp_results = pd.merge(sessions, lfp_results, how='left')
In [9]:
if stim_loc == 'ms':
lfp_results = lfp_results.query('stim_location!="mecl" and stim_location!="mecr"')
elif stim_loc == 'mec':
lfp_results = lfp_results.query('stim_location!="ms"')
else:
raise AssertionError('')
In [10]:
def action_group(row):
a = int(row.channel_group in [0,1,2,3])
return f'{row.action}-{a}'
lfp_results['action_side_a'] = lfp_results.apply(action_group, axis=1)
In [11]:
lfp_results['stim_strength'] = lfp_results['stim_p_max'] / lfp_results['theta_bandpower']
In [12]:
# lfp_results_hemisphere = lfp_results.sort_values(
# by=['action_side_a', 'stim_strength', 'signal_to_noise'], ascending=[True, False, False]
lfp_results_hemisphere = lfp_results.sort_values(
by=['action_side_a', 'channel_group'], ascending=[True, False]
).drop_duplicates(subset='action_side_a', keep='first')
lfp_results_hemisphere.loc[:,['action_side_a','channel_group', 'signal_to_noise', 'stim_strength']].head()
Out[12]:
In [13]:
colors = ['#1b9e77','#d95f02','#7570b3','#e7298a']
labels = ['Baseline I', '11 Hz', 'Baseline II', '30 Hz']
# Hz11 means that the baseline session was indeed before an 11 Hz session
queries = ['baseline and i and Hz11', 'frequency==11', 'baseline and ii and Hz30', 'frequency==30']
In [14]:
# prepare pairwise comparison: same animal same side same date different sessions
In [15]:
def make_entity_date_side(row):
s = row.action_side_a.split('-')
del s[2]
return '-'.join(s)
In [16]:
lfp_results_hemisphere['entity_date_side'] = lfp_results_hemisphere.apply(make_entity_date_side, axis=1)
In [17]:
from functools import reduce
In [18]:
keys = [
'theta_bandpower',
'theta_relpower',
'theta_relpeak',
'theta_peak',
'theta_freq',
'theta_half_width',
'stim_bandpower',
'stim_relpower',
'stim_relpeak',
'stim_half_width',
'stim_p_max',
'stim_strength',
]
results = {}
for key in keys:
results[key] = list()
for query, label in zip(queries, labels):
values = lfp_results_hemisphere.query(query).loc[:,['entity_date_side', key]]
results[key].append(values.rename({key: label}, axis=1))
for key, val in results.items():
df = reduce(lambda left,right: pd.merge(left, right, on='entity_date_side', how='outer'), val)
results[key] = df.drop('entity_date_side', axis=1)
In [19]:
xlabel = {
'theta_bandpower': 'Theta bandpower (V$^2$/Hz)',
'theta_relpower': 'Theta relative power',
'theta_relpeak': 'Theta relative power',
'theta_peak': 'Peak PSD (V$^2$/Hz)',
'theta_freq': '(Hz)',
'theta_half_width': '(Hz)',
}
for key in xlabel:
fig = plt.figure(figsize=(3.5,2))
plt.suptitle(key)
legend_lines = []
for color, label in zip(colors, labels):
legend_lines.append(matplotlib.lines.Line2D([0], [0], color=color, label=label))
sns.kdeplot(data=results[key].loc[:,labels], cumulative=True, legend=False, palette=colors, common_norm=False)
plt.legend(
handles=legend_lines,
bbox_to_anchor=(1.04,1), borderaxespad=0, frameon=False)
plt.tight_layout()
plt.grid(False)
despine()
plt.xlabel(xlabel[key])
figname = f'lfp-psd-histogram-{key}'
fig.savefig(
output_path / 'figures' / f'{figname}.png',
bbox_inches='tight', transparent=True)
fig.savefig(
output_path / 'figures' / f'{figname}.svg',
bbox_inches='tight', transparent=True)
In [20]:
xlabel = {
'stim_bandpower': 'Energy (V$^2$/Hz)',
'stim_relpower': 'Relative power',
'stim_relpeak': 'Relative power',
'stim_half_width': '(Hz)',
'stim_p_max': 'Peak PSD (V$^2$/Hz)',
'stim_strength': 'Ratio',
}
for key in xlabel:
fig = plt.figure(figsize=(3.2,2))
plt.suptitle(key)
legend_lines = []
for color, label in zip(colors[1::2], labels[1::2]):
legend_lines.append(matplotlib.lines.Line2D([0], [0], color=color, label=label))
sns.kdeplot(data=results[key].loc[:, labels[1::2]], cumulative=True, legend=False, palette=colors[1::2], common_norm=False)
plt.legend(
handles=legend_lines,
bbox_to_anchor=(1.04,1), borderaxespad=0, frameon=False)
plt.tight_layout()
plt.grid(False)
despine()
plt.xlabel(xlabel[key])
figname = f'lfp-psd-histogram-{key}'
fig.savefig(
output_path / 'figures' / f'{figname}.png',
bbox_inches='tight', transparent=True)
fig.savefig(
output_path / 'figures' / f'{figname}.svg',
bbox_inches='tight', transparent=True)
In [21]:
def summarize(data):
return "{:.1e} ± {:.1e} ({})".format(data.mean(), data.sem(), sum(~np.isnan(data)))
def MWU(df, keys):
'''
Mann Whitney U
'''
Uvalue, pvalue = scipy.stats.mannwhitneyu(
df[keys[0]].dropna(),
df[keys[1]].dropna(),
alternative='two-sided')
return "{:.2f}, {:.3f}".format(Uvalue, pvalue)
def PRS(df, keys):
'''
Permutation ReSampling
'''
pvalue, observed_diff, diffs = permutation_resampling(
df[keys[0]].dropna(),
df[keys[1]].dropna(), statistic=np.median)
return "{:.2f}, {:.3f}".format(observed_diff, pvalue)
def wilcoxon(df, keys):
dff = df.loc[:,[keys[0], keys[1]]].dropna()
statistic, pvalue = scipy.stats.wilcoxon(
dff[keys[0]],
dff[keys[1]],
alternative='two-sided')
return "{:.1e}, {:.1e}, ({})".format(statistic, pvalue, len(dff))
def summarize_wilcoxon(df, keys):
dff = df.loc[:,[keys[0], keys[1]]].dropna()
return"{:.1e} ± {:.1e}, {:.1e} ± {:.1e} ({})".format(dff[keys[0]].mean(), dff[keys[0]].sem(), dff[keys[1]].mean(), dff[keys[1]].sem(), len(dff))
def paired_t(df, keys):
dff = df.loc[:,[keys[0], keys[1]]].dropna()
statistic, pvalue = scipy.stats.ttest_rel(
dff[keys[0]],
dff[keys[1]])
return "{:.2f}, {:.3f}".format(statistic, pvalue)
def normality(df, key):
statistic, pvalue = scipy.stats.normaltest(
df[key].dropna())
return "{:.1e}, {:.1e}".format(statistic, pvalue)
def shapiro(df, key):
statistic, pvalue = scipy.stats.shapiro(
df[key].dropna())
return "{:.2f}, {:.3f}".format(statistic, pvalue)
def rename(name):
return name.replace("_field", "-field").replace("_", " ").capitalize()
In [22]:
stat = pd.DataFrame()
for key, df in results.items():
Key = rename(key)
stat[Key] = df.agg(summarize)
stat[Key] = df.agg(summarize)
for i, c1 in enumerate(df.columns):
try:
stat.loc[f'Normality {c1}', Key] = normality(df, c1)
except:
stat.loc[f'Normality {c1}', Key] = np.nan
# stat.loc[f'Shapiro {c1}', Key] = shapiro(df, c1)
for c2 in df.columns[i+1:]:
# stat.loc[f'MWU {c1} {c2}', Key] = MWU(df, [c1, c2])
# stat.loc[f'PRS {c1} {c2}', Key] = PRS(df, [c1, c2])
try:
stat.loc[f'Wilcoxon {c1} {c2}', Key] = wilcoxon(df, [c1, c2])
except:
stat.loc[f'Wilcoxon {c1} {c2}', Key] = np.nan
# stat.loc[f'Paired T {c1} {c2}', Key] = paired_t(df, [c1, c2])
stat.loc[f'Paired summary {c1} {c2}', Key] = summarize_wilcoxon(df, [c1, c2])
stat.sort_index()
Out[22]:
In [ ]:
stat.to_latex(output_path / "statistics" / f"statistics.tex")
stat.to_csv(output_path / "statistics" / f"statistics.csv")
In [23]:
for key, result in results.items():
result.to_latex(output_path / "statistics" / f"values_{key}.tex")
result.to_csv(output_path / "statistics" / f"values_{key}.csv")
Plot PSD¶
In [24]:
psd = pd.read_feather(pathlib.Path("output") / ("stimulus-lfp-response" + zscore_str) / 'data' / 'psd.feather')
freqs = pd.read_feather(pathlib.Path("output") / ("stimulus-lfp-response" + zscore_str) / 'data' / 'freqs.feather')
In [25]:
from septum_mec.analysis.plotting import plot_bootstrap_timeseries
In [26]:
freq = freqs.T.iloc[0].values
mask = (freq < 49)
In [27]:
fig, axs = plt.subplots(1, 2, sharex=True, sharey=True, figsize=(5,2))
axs = axs.repeat(2)
for i, (ax, query) in enumerate(zip(axs.ravel(), queries)):
selection = [
f'{r.action}_{r.channel_group}'
for i, r in lfp_results_hemisphere.query(query).iterrows()]
values = psd.loc[mask, selection].to_numpy()
values = 10 * np.log10(values)
plot_bootstrap_timeseries(freq[mask], values, ax=ax, lw=1, label=labels[i], color=colors[i])
# ax.set_title(titles[i])
ax.set_xlabel('Frequency Hz')
ax.legend(frameon=False)
axs[0].set_ylabel('PSD (dB/Hz)')
# axs[0].set_ylim(-31, 1)
despine()
figname = 'lfp-psd'
fig.savefig(
output_path / 'figures' / f'{figname}.png',
bbox_inches='tight', transparent=True)
fig.savefig(
output_path / 'figures' / f'{figname}.svg',
bbox_inches='tight', transparent=True)
Store results in Expipe action¶
In [28]:
action = project.require_action("stimulus-lfp-response" + '-' + stim_loc + zscore_str)
In [29]:
copy_tree(output_path, str(action.data_path()))
Out[29]:
In [30]:
septum_mec.analysis.registration.store_notebook(action, "20_stimulus-lfp-response.ipynb")
In [ ]: