Develop an analysis workflow¶
Goal:¶
build an analysis workflow centered on database interactions
Requirements:¶
- completion of all previous tutorials
Contents:¶
- create a summary visualization of the dataset
- create a summary visualization of a session's spiking activity and movie contents
- correlate binned character appearances and spiking activity
from collections import Counter
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from epiphyte.database.db_setup import *
from epiphyte.database.query_functions import *
from epiphyte.preprocessing.data_preprocessing.binning import *
erd = dj.ERD(epi_schema)
erd
Create a summary visualization of all units in the database¶
- collect single- and multi-unit data from all sessions
- collect region-wise information for all units within each session
- visualize the watchlogs of each patient
- visualize the annotations on the base version of the movie (no specific pauses or skips factored in)
su_count = len(ElectrodeData & "unit_type='S'")
mu_count = len(ElectrodeData & "unit_type='M'")
region_df = pd.DataFrame(
{ "patient_id": ElectrodeData.fetch("patient_id"),
"session_nr": ElectrodeData.fetch("session_nr"),
"region": ElectrodeData.fetch("brain_region")}
)
counts = (
region_df
.groupby(["patient_id", "session_nr", "region"])
.size()
.reset_index(name="count")
)
counts["patient_session"] = (
counts["patient_id"].astype(str) + "_s" + counts["session_nr"].astype(str)
)
pivot_df = counts.pivot(
index="patient_session",
columns="region",
values="count"
).fillna(0)
annotations_df = pd.DataFrame(MovieAnnotation())
mosaic = """
ABB
CDD
"""
fig = plt.figure(figsize=(6,3),)
axes = fig.subplot_mosaic(
mosaic,
gridspec_kw={
"height_ratios": [2, 1.5]
}
)
######
ax = axes["A"]
bar_colors = sns.color_palette("viridis", n_colors=2)
ax.bar([0, 1], [su_count, mu_count], color=bar_colors)
ax.text(0-0.15, su_count+3, f"{su_count}")
ax.text(1-0.15, mu_count+3, f"{mu_count}")
ax.set_xticks([0,1])
ax.set_xticklabels(["SU", "MU"])
ax.set_ylabel("Count")
ax.set_title("Unit Counts")
sns.despine(ax=ax)
######
ax = axes["B"]
pivot_df.plot(
ax=ax,
kind="bar",
stacked=True,
figsize=(10,6),
colormap="viridis"
)
ax.legend(frameon=False, bbox_transform=ax.transAxes, bbox_to_anchor = [1,1])
ax.set_title("Region Distribution by Patient & Session")
ax.set_xlabel("Patient-Session")
ax.set_ylabel("Count")
ax.tick_params(axis='x', rotation=45)
sns.despine(ax=ax)
######
ax = axes["C"]
pats, seshs = MovieSession.fetch("patient_id", "session_nr")
colors = sns.color_palette("viridis_r", n_colors=len(pats))
i = 0
for pat, sesh in zip(pats, seshs):
pts, dts = (MovieSession & f"patient_id={pat}" & f"session_nr={sesh}").fetch("pts", "dts")
pts = pts[0]
dts = dts[0] - dts[0][0]
dts = dts / 1000
ax.plot(dts, pts, color=colors[i], label=f"{pat}-s{sesh}")
i += 1
ax.set_title("Patient Watchlogs")
ax.set_xlabel("Time [s]")
ax.set_ylabel("Frame index")
sns.despine(ax=ax)
ax.legend(frameon=False)
######
ax = axes["D"]
yticks = []
ylabels = []
y_pos = 0
bar_colors = sns.color_palette("viridis", n_colors=3)
bar_height = 0.8
for idx, row in annotations_df.iterrows():
starts = row['start_times']
stops = row['stop_times']
values = row['values']
color = bar_colors[idx]
label = row['label_name']
for start, stop, value in zip(starts, stops, values):
if value:
ax.axvspan(
xmin=start,
xmax=stop,
ymin=y_pos / len(annotations_df),
ymax=(y_pos + bar_height) / len(annotations_df),
facecolor=color,
alpha=1.0,
edgecolor='white'
)
yticks.append(y_pos + bar_height / 2)
ylabels.append(label)
y_pos += 1
ax.set_ylim(0, len(annotations_df))
ax.set_yticks(yticks)
ax.set_yticklabels(ylabels)
ax.set_xlabel("Time [s]")
ax.set_title("Annotations")
sns.despine(ax=ax)
plt.tight_layout()
plt.show()
Visualize stimulus events and neuronal activity¶
- collect the patient-aligned labels (i.e. those with pauses and any skips in playback factored into the PTS vector)
- collect all spike times recorded from patient 1, session 1
- overlay the annotations on top of the spikes
- use preset query functions to simplify database calls
patient_id = 1
session_nr = 1
patient_spikes = (SpikeData & f"patient_id={patient_id}" & f"session_nr={session_nr}").fetch("spike_times")
rectime = get_patient_neural_rectime(patient_id, session_nr)
fig, ax = plt.subplots(1,1,figsize=(14, 6))
ax.eventplot(patient_spikes, lw=0.2, color="black")
colors = sns.color_palette("viridis", n_colors=3)
for i, label_name in enumerate(MovieAnnotation.fetch("label_name")):
values, starts, stops = get_patient_aligned_label_info(patient_id, session_nr, label_name)
c = 0
for start, stop, value in zip(starts, stops, values):
if c == 0:
label_name = label_name
else:
label_name = None
if value:
ax.axvspan(
xmin=start,
xmax=stop,
ymin=0,
ymax=len(patient_spikes)-5,
facecolor=colors[i],
alpha=0.4,
edgecolor='white',
label=label_name
)
c += 1
ax.set_xticks(rectime[::10_000])
ax.set_xticklabels(((rectime[::10_000] - rectime[0]) / 1000).round(2))
ax.set_yticks([])
ax.set_xlabel("Time [s]\nAligned to recording onset")
ax.set_ylabel("Neurons")
ax.set_title("Patient 1 spikes + aligned movie labels")
ax.legend(frameon=False, bbox_transform=ax.transAxes, bbox_to_anchor = [1.1,1])
sns.despine(left=True)
plt.show()
Calculate the correlation between character appearance and spiking activity¶
- collect label and spiking information
- bin label to 80 ms bins, excluding any pauses
- bin spikes to 80 ms bins, also excluding any pause
- correlate the resulting binned character presence and spiking activity
Note: spiking activity is totally random.
label_name = "character1"
values, starts, stops = get_patient_aligned_label_info(patient_id, session_nr, label_name)
spike_times = get_spike_times(patient_id, session_nr, 1)
bin_size = 80
label_binned = bin_label(
patient_id=1,
session_nr=1,
values=values,
start_times=starts,
stop_times=stops,
bin_size=bin_size,
exclude_pauses=True,
)
spikes_binned = bin_spikes(
patient_id=1,
session_nr=1,
spike_times=spike_times,
bin_size=bin_size,
exclude_pauses=True,
output_edges=False,)
np.corrcoef(label_binned, spikes_binned)[0,1]
np.float64(-0.0015086612955703923)