Contents

R2 in the cerebellum

This analysis is supplementary to the results reported in R2 in the cortex. The difference is that here the summary statistics are computed in a mask including the cerebellum, instead of the cortex.

import os
import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns
from cneuromod_embeddings.r2_summary import _r2_intra, _r2_inter, _r2_other

sns.set_theme(style="whitegrid")
sns.set(font_scale=1.5)

R2 with fwhm=5

Overall, with fwhm=5, the results are very similar to what was observed in the cortex, except the R2 values are lower (by about 0.1).

DYPAC intra vs DIUFMO

path_data = '/home/pbellec/git/cneuromod_embeddings/cneuromod_embeddings/friends-s01/'
path_results = os.path.join(path_data, 'r2_friends-s01_cerebellum')

fwhm = 5

# DYPAC
dypac60  = os.path.join(path_results, f'r2_fwhm-intra_fwhm-{fwhm}_cluster-20_state-60.p')
dypac120 = os.path.join(path_results, f'r2_fwhm-intra_fwhm-{fwhm}_cluster-20_state-120.p')
dypac150 = os.path.join(path_results, f'r2_fwhm-intra_fwhm-{fwhm}_cluster-50_state-150.p')
dypac300 = os.path.join(path_results, f'r2_fwhm-intra_fwhm-{fwhm}_cluster-50_state-300.p')
dypac900 = os.path.join(path_results, f'r2_fwhm-intra_fwhm-{fwhm}_cluster-300_state-900.p')

# DYPAC inter-subject
inter60  = os.path.join(path_results, f'r2_fwhm-inter_fwhm-{fwhm}_cluster-20_state-60.p')
inter120 = os.path.join(path_results, f'r2_fwhm-inter_fwhm-{fwhm}_cluster-20_state-120.p')
inter150 = os.path.join(path_results, f'r2_fwhm-inter_fwhm-{fwhm}_cluster-50_state-150.p')
inter300 = os.path.join(path_results, f'r2_fwhm-inter_fwhm-{fwhm}_cluster-50_state-300.p')
inter900 = os.path.join(path_results, f'r2_fwhm-inter_fwhm-{fwhm}_cluster-300_state-900.p')

# DIFUMO
difumo256  = os.path.join(path_results, f'r2_fwhm-difumo256_fwhm-{fwhm}.p')
difumo512  = os.path.join(path_results, f'r2_fwhm-difumo512_fwhm-{fwhm}.p')
difumo1024 = os.path.join(path_results, f'r2_fwhm-difumo1024_fwhm-{fwhm}.p')

# MIST
mist197 = os.path.join(path_results, f'r2_fwhm-mist197_fwhm-{fwhm}.p')
mist444 = os.path.join(path_results, f'r2_fwhm-mist444_fwhm-{fwhm}.p')

# Schaefer
schaefer = os.path.join(path_results, f'r2_fwhm-schaefer_fwhm-{fwhm}.p')

# Smith
smith70 = os.path.join(path_results, f'r2_fwhm-smith_fwhm-{fwhm}.p')

val_r2 = pd.read_pickle(difumo256)
val_r2 = val_r2.append(pd.read_pickle(difumo512))
val_r2 = val_r2.append(pd.read_pickle(difumo1024))
val_r2 = val_r2.append(pd.read_pickle(dypac900))

fig = plt.figure(figsize=(20, 15))
sns.boxenplot(data=val_r2, x='subject', y='r2', hue='params', scale='area')
plt.ylabel('R2 embedding quality')
plt.title(f'FWHM={fwhm}')

Text(0.5, 1.0, 'FWHM=5')
_images/friends-s01_r2_cerebellum_4_1.png

DYPAC intra vs other group parcellations

val_r2 = pd.read_pickle(smith70)
val_r2 = val_r2.append(pd.read_pickle(mist197))
val_r2 = val_r2.append(pd.read_pickle(mist444))
val_r2 = val_r2.append(pd.read_pickle(schaefer))
val_r2 = val_r2.append(pd.read_pickle(dypac900))

fig = plt.figure(figsize=(20, 15))
sns.boxenplot(data=val_r2, x='subject', y='r2', hue='params', scale='area')
plt.ylabel('R2 embedding quality')
plt.title(f'FWHM={fwhm}')

Text(0.5, 1.0, 'FWHM=5')
_images/friends-s01_r2_cerebellum_6_1.png

DYPAC multi-resolution

val_r2 = pd.read_pickle(dypac60)
val_r2 = val_r2.append(pd.read_pickle(dypac120))
val_r2 = val_r2.append(pd.read_pickle(dypac150))
val_r2 = val_r2.append(pd.read_pickle(dypac300))
val_r2 = val_r2.append(pd.read_pickle(dypac900))

fig = plt.figure(figsize=(20, 15))
sns.boxenplot(data=val_r2, x='subject', y='r2', hue='params', scale='area')
plt.ylabel('R2 embedding quality')
plt.title(f'FWHM={fwhm}')

Text(0.5, 1.0, 'FWHM=5')
_images/friends-s01_r2_cerebellum_8_1.png

intra vs inter subject R2

val_r2 = pd.read_pickle(dypac60)
val_r2 = val_r2.append(pd.read_pickle(dypac120))
val_r2 = val_r2.append(pd.read_pickle(dypac150))
val_r2 = val_r2.append(pd.read_pickle(dypac300))
val_r2 = val_r2.append(pd.read_pickle(dypac900))
val_r2 = val_r2.append(pd.read_pickle(inter60))
val_r2 = val_r2.append(pd.read_pickle(inter120))
val_r2 = val_r2.append(pd.read_pickle(inter150))
val_r2 = val_r2.append(pd.read_pickle(inter300))
val_r2 = val_r2.append(pd.read_pickle(inter900))

fig = plt.figure(figsize=(20, 15))
sns.boxenplot(data=val_r2, x='params', y='r2', hue='type', scale='area')
plt.ylabel('R2 embedding quality')
plt.title(f'FWHM={fwhm}')

Text(0.5, 1.0, 'FWHM=5')
_images/friends-s01_r2_cerebellum_10_1.png

R2 with fwhm=8

Overall, with fwhm=8, the results are very similar to what was observed in the cortex. Unlike fwhm=5, the R2 values are quite similar in the cerebellum and the cortex. The only difference is that the difference between DYPAC and DIFUMO is more pronounced in the cerebellum (i.e. DIFUMO does not perform as well in the cerebellum as in the cortex for fwhm=8), and the improvement in R2 for cluster-300_state-900 is more important in the cerebellum than in the cortex.

path_results = '/data/cisl/pbellec/cneuromod_embeddings/xp_202012/r2_friends-s01_cerebellum/'
fwhm = 8

# DYPAC
dypac60  = os.path.join(path_results, f'r2_fwhm-intra_fwhm-{fwhm}_cluster-20_state-60.p')
dypac120 = os.path.join(path_results, f'r2_fwhm-intra_fwhm-{fwhm}_cluster-20_state-120.p')
dypac150 = os.path.join(path_results, f'r2_fwhm-intra_fwhm-{fwhm}_cluster-50_state-150.p')
dypac300 = os.path.join(path_results, f'r2_fwhm-intra_fwhm-{fwhm}_cluster-50_state-300.p')
dypac900 = os.path.join(path_results, f'r2_fwhm-intra_fwhm-{fwhm}_cluster-300_state-900.p')

# DYPAC inter-subject
inter60  = os.path.join(path_results, f'r2_fwhm-inter_fwhm-{fwhm}_cluster-20_state-60.p')
inter120 = os.path.join(path_results, f'r2_fwhm-inter_fwhm-{fwhm}_cluster-20_state-120.p')
inter150 = os.path.join(path_results, f'r2_fwhm-inter_fwhm-{fwhm}_cluster-50_state-150.p')
inter300 = os.path.join(path_results, f'r2_fwhm-inter_fwhm-{fwhm}_cluster-50_state-300.p')
inter900 = os.path.join(path_results, f'r2_fwhm-inter_fwhm-{fwhm}_cluster-300_state-900.p')

# DIFUMO
difumo256  = os.path.join(path_results, f'r2_fwhm-difumo256_fwhm-{fwhm}.p')
difumo512  = os.path.join(path_results, f'r2_fwhm-difumo512_fwhm-{fwhm}.p')
difumo1024 = os.path.join(path_results, f'r2_fwhm-difumo1024_fwhm-{fwhm}.p')

# MIST
mist197 = os.path.join(path_results, f'r2_fwhm-mist197_fwhm-{fwhm}.p')
mist444 = os.path.join(path_results, f'r2_fwhm-mist444_fwhm-{fwhm}.p')

# Schaefer
schaefer = os.path.join(path_results, f'r2_fwhm-schaefer_fwhm-{fwhm}.p')

# Smith
smith70 = os.path.join(path_results, f'r2_fwhm-smith_fwhm-{fwhm}.p')

DYPAC intra vs DIFUMO

val_r2 = pd.read_pickle(difumo256)
val_r2 = val_r2.append(pd.read_pickle(difumo512))
val_r2 = val_r2.append(pd.read_pickle(difumo1024))
val_r2 = val_r2.append(pd.read_pickle(dypac900))

fig = plt.figure(figsize=(20, 15))
sns.boxenplot(data=val_r2, x='subject', y='r2', hue='params', scale='area')
plt.ylabel('R2 embedding quality')
plt.title(f'FWHM={fwhm}')

Text(0.5, 1.0, 'FWHM=8')
_images/friends-s01_r2_cerebellum_14_1.png

DYPAC intra vs other group parcellations

val_r2 = pd.read_pickle(smith70)
val_r2 = val_r2.append(pd.read_pickle(mist197))
val_r2 = val_r2.append(pd.read_pickle(mist444))
val_r2 = val_r2.append(pd.read_pickle(schaefer))
val_r2 = val_r2.append(pd.read_pickle(dypac900))

fig = plt.figure(figsize=(20, 15))
sns.boxenplot(data=val_r2, x='subject', y='r2', hue='params', scale='area')
plt.ylabel('R2 embedding quality')
plt.title(f'FWHM={fwhm}')

Text(0.5, 1.0, 'FWHM=8')
_images/friends-s01_r2_cerebellum_16_1.png

DYPAC multi-resolution

val_r2 = pd.read_pickle(dypac60)
val_r2 = val_r2.append(pd.read_pickle(dypac120))
val_r2 = val_r2.append(pd.read_pickle(dypac150))
val_r2 = val_r2.append(pd.read_pickle(dypac300))
val_r2 = val_r2.append(pd.read_pickle(dypac900))

fig = plt.figure(figsize=(20, 15))
sns.boxenplot(data=val_r2, x='subject', y='r2', hue='params', scale='area')
plt.ylabel('R2 embedding quality')
plt.title(f'FWHM={fwhm}')

Text(0.5, 1.0, 'FWHM=8')
_images/friends-s01_r2_cerebellum_18_1.png

Intra vs inter subject R2

val_r2 = pd.read_pickle(dypac60)
val_r2 = val_r2.append(pd.read_pickle(dypac120))
val_r2 = val_r2.append(pd.read_pickle(dypac150))
val_r2 = val_r2.append(pd.read_pickle(dypac300))
val_r2 = val_r2.append(pd.read_pickle(dypac900))
val_r2 = val_r2.append(pd.read_pickle(inter60))
val_r2 = val_r2.append(pd.read_pickle(inter120))
val_r2 = val_r2.append(pd.read_pickle(inter150))
val_r2 = val_r2.append(pd.read_pickle(inter300))
val_r2 = val_r2.append(pd.read_pickle(inter900))

fig = plt.figure(figsize=(20, 15))
sns.boxenplot(data=val_r2, x='params', y='r2', hue='type', scale='area')
plt.ylabel('R2 embedding quality')
plt.title(f'FWHM={fwhm}')

Text(0.5, 1.0, 'FWHM=8')
_images/friends-s01_r2_cerebellum_20_1.png
R2 in central structures