def _parallel_nbs(self, FreqBand, thresh):
"""
Calculates nbs for one Frequency band. Is called in calc_nbs.
:return dataframe with pvals of
"""
print(f'Processing {FreqBand}, Thresh: {thresh}')
ResultDict = {
'Freq': [],
'Threshold': [],
'P-Val': [],
'Component-File': [],
'Index': []
}
GroupFCList = []
for Group in self.GroupIDs.keys():
# Find File and load file
GroupFCs = np.load(
self.find(suffix='stacked-FCs',
filetype='.npy',
Group=Group,
Freq=FreqBand))
# Transform matrix for nbs (NxNxP with P being the number of subjects per group)
tGroupFCs = np.moveaxis(GroupFCs, 0, -1)
GroupFCList.append(tGroupFCs)
# Set Component File Path
CompFileName = self.createFileName(suffix='Component-Adj',
filetype='.npy',
Freq=FreqBand,
Thresh=thresh)
CompFilePath = self.createFilePath(self.EdgeStatsDir, 'NetBasedStats',
'Components', CompFileName)
pval, adj, null = nbs_bct(GroupFCList[0],
GroupFCList[1],
thresh=thresh,
k=1000)
print('Adjacency Shape: ', adj.shape)
for idx, p in enumerate(pval):
ResultDict['Freq'].append(FreqBand)
ResultDict['Threshold'].append(thresh)
ResultDict['P-Val'].append(p)
ResultDict['Component-File'].append(CompFilePath)
ResultDict['Index'].append(idx + 1)
# Save Null-Sample and Component to File
NullFileName = self.createFileName(suffix='Null-Sample',
filetype='.npy',
Freq=FreqBand,
Thresh=thresh)
NullFilePath = self.createFilePath(self.EdgeStatsDir, 'NetBasedStats',
'Null-Sample', NullFileName)
np.save(NullFilePath, null)
np.save(CompFilePath, adj)
# Return dataframe
df = pd.DataFrame(ResultDict, index=ResultDict['Index'])
return df
def apply_nbs(X, y, th, iterations=1000, verbose=False):
"""
Application of NBS. Computation of max component size, its null
distribution and p-value.
Parameters:
----------
X: shape = ((N+M), n, n), where N and M are the sizes of the two grous, and
n is the number of nodes in the graphs.
Data from both groups. (N+M) connectivity matrices.
y: shape=(N+M,)
Class labels for the data
th: float
Minimum t-value used as threshold
iterations: int
Number of permutations used to estimate the empirical null distribution
verbose: bool
Verbosity
Returns:
-------
max_comp_size: float
The size of the biggest connected component.
max_comp_null: array
Null distribution of the size of biggest connected component.
p_value: float
p-value
"""
#taking the two groups
XX = X[y == 0].T
YY = X[y == 1].T
#applying nbs
pval, adj, max_comp_null = nbs.nbs_bct(XX,
YY,
th,
iterations,
verbose=verbose)
#Computing connected component sizes
_, comp_sizes = np.unique(adj[adj > 0], return_counts=True)
max_comp_size = np.max(
comp_sizes) / 2 #div by 2 because edges are counted twice
return max_comp_size, max_comp_null, np.min(pval)
def apply_nbs(X, y, th, iterations=1000, verbose=False):
"""
Application of NBS. Computation of max component size, its null
distribution and p-value.
Parameters:
----------
X: shape = ((N+M), n, n), where N and M are the sizes of the two grous, and
n is the number of nodes in the graphs.
Data from both groups. (N+M) connectivity matrices.
y: shape=(N+M,)
Class labels for the data
th: float
Minimum t-value used as threshold
iterations: int
Number of permutations used to estimate the empirical null distribution
verbose: bool
Verbosity
Returns:
-------
max_comp_size: float
The size of the biggest connected component.
max_comp_null: array
Null distribution of the size of biggest connected component.
p_value: float
p-value
"""
#taking the two groups
XX = X[y == 0].T
YY = X[y == 1].T
#applying nbs
pval, adj, max_comp_null = nbs.nbs_bct(XX, YY, th, iterations,
verbose=verbose)
#Computing connected component sizes
_, comp_sizes = np.unique(adj[adj>0], return_counts=True)
max_comp_size = np.max(comp_sizes) / 2 #div by 2 because edges are counted twice
return max_comp_size, max_comp_null, np.min(pval)
reshape_Ses2_trauma = np.moveaxis(np.array(cor_z_array_2),0,-1)
#%% create a symmetric matrix for CPM
sym_mat = []
for i in range(len(ses_1_trauma_corr_z)):
x = nilearn.connectome.sym_matrix_to_vec(ses_1_trauma_corr_z[i], discard_diagonal=False)
x_mat = nilearn.connectome.vec_to_sym_matrix(x)
sym_mat.append(x_mat)
sym_mat = np.array(sym_mat)
sym_mat = np.moveaxis(sym_mat,0,-1)
scipy.io.savemat('ses2_trauma.mat', dict(x=sym_mat))
#%% run NBS
from bct import nbs
# we compare ket1 and ket3
pval, adj, _ = nbs.nbs_bct(reshape_ses1_trauma, reshape_Ses2_trauma, thresh=2.5, tail='both',k=500, paired=True, verbose = True)
# there is a difference. Lets plot the network
# first create diff
diffMat_2_1 = contFuncs(ses_1_trauma_corr, ses_2_trauma_corr)
# reshape and save as .mat for CPM
diffMat_21_CPM = np.moveaxis(np.array(diffMat_2_1),0, -1)
scipy.io.savemat('diffMat_2_1.mat', dict(x=diffMat_21_CPM))
diffMat_2_1_thr = np.array(diffMat_2_1) * adj
diffMat_2_1_thr_avergae = np.mean(diffMat_2_1_thr, axis = 0)
# how many edges?
np.sum(adj)
# 1310 edges survived
np.savetxt('diffMat_2_1_thr.csv', diffMat_2_1_thr_avergae)
ket4Reshape = np.moveaxis(np.array(ket4_corr), 0,-1) mid1Reshape = np.moveaxis(np.array(mid1_corr),0,-1) mid2Reshape = np.moveaxis(np.array(mid2_corr),0,-1) mid3Reshape = np.moveaxis(np.array(mid3_corr),0,-1) #print(mid3Reshape.shape) # In[ ]: # now we can run NBS # NBS is taken from: https://github.com/aestrivex/bctpy, can be installed using pip (pip install bctpy) from bct import nbs # we compare ket1 and ket3 pval, adj, _ = nbs.nbs_bct(ket1Reshape, ket3Reshape, thresh=3, tail='both',k=1000, paired=True, verbose = True) # check mean p vlue #np.mean(checkNBS[0]) # In[ ]: # look at p values and No. of components. print(pval.shape) print (pval) len(pval) print(adj.shape) print(adj[0:10]) ad = np.array(adj) print(ad[:,0:10]) #bct.adjacency_plot_und(adj, coords, tube=False)
# %%
# lets run NBS
ketDeltaReshape = np.moveaxis(np.array(ketDelta), 0, -1)
midDeltaReshape = np.moveaxis(np.array(midDelta), 0, -1)
ketSes2_reshape = np.moveaxis(np.array(ketSes2), 0, -1)
midSes2_reshape = np.moveaxis(np.array(midSes2), 0, -1)
print(ketDeltaReshape.shape)
print(midDeltaReshape.shape)
from bct import nbs
# we compare ket1 and ket3
pval, adj, _ = nbs.nbs_bct(ketDeltaReshape,
midDeltaReshape,
thresh=2.3,
tail='both',
k=1000,
paired=False,
verbose=False)
print(pval)
# %%
# ok lets threshold using adjacency
#tTresh = t[np.tril(adj)]
tTresh = t * adj
#tTresh[np.triu(tTresh)] = t
sns.heatmap(tTresh,
xticklabels=labels,
yticklabels=labels,
cmap='coolwarm',
annot=True)
ket4Reshape = np.moveaxis(np.array(ket4_corr), 0,-1) mid1Reshape = np.moveaxis(np.array(mid1_corr),0,-1) mid2Reshape = np.moveaxis(np.array(mid2_corr),0,-1) mid3Reshape = np.moveaxis(np.array(mid3_corr),0,-1) #print(mid3Reshape.shape) # In[ ]: # now we can run NBS # NBS is taken from: https://github.com/aestrivex/bctpy, can be installed using pip (pip install bctpy) from bct import nbs # we compare ket1 and ket3 pval, adj, _ = nbs.nbs_bct(ket1Reshape, ket3Reshape, thresh=3, tail='both',k=1000, paired=True, verbose = True) # check mean p vlue #np.mean(checkNBS[0]) # In[ ]: # look at p values and No. of components. print(pval.shape) print (pval) len(pval) print(adj.shape) print(adj[0:10]) ad = np.array(adj) print(ad[:,0:10]) #bct.adjacency_plot_und(adj, coords, tube=False)
#%%
#%% create time series (with condounders)
session1 = timeSeries(func_files=fileList(subject_list_1,'1'), confound_files=confList(subject_list_1, '1'))
session2 = timeSeries(func_files=fileList(subject_list_1,'2'), confound_files=confList(subject_list_1, '2'))
session3 = timeSeries(func_files=fileList(subject_list3,'3'), confound_files=confList(subject_list3, '3'))
#%%
os.chdir('/home/or/kpe_conn/ShenParc')
np.save("session_1Timeseries_ShenRS",session1) # saving array
np.save("session_2TimeseriesShenRS",session2)
np.save("session_3TimeseriesShenRS", session3)
#%% Correlations
cor1 = createCorMat(session1)
cor2 = createCorMat(session2)
cor3 = createCorMat(session3)
#%% NBS
cor1Reshape = np.moveaxis(np.array(cor1),0,-1)
cor2Reshape = np.moveaxis(np.array(cor2),0,-1)
cor3Reshape = np.moveaxis(np.array(cor3),0,-1)
from bct import nbs
# we compare ket1 and ket3
pval, adj, _ = nbs.nbs_bct(cor1Reshape, cor2Reshape, thresh=2.5, tail='both',k=500, paired=True, verbose = True)
# no difference in RS across groups
# Compare first and 3rd
pval, adj, _ = nbs.nbs_bct(cor1Reshape, cor2Reshape, thresh=2.5, tail='both',k=500, paired=True, verbose = True)
plotting.plot_connectome(empty, coords, edge_threshold='95%', colorbar=True, black_bg = False, annotate = True, node_color = color_node) # plot in browser view = plotting.view_connectome(empty, coords, threshold='90%') view.open_in_browser() view_color.open_in_browser() #%% Run Network Based Analysis # first reshape the matrix dimensions (right now its [subs,x,y]) to [x,y,subs] trt1Reshape = np.moveaxis(np.array(trauma_1st_ses),0,-1) trt2Reshape = np.moveaxis(np.array(trauma_2_1st_ses),0,-1) from bct import nbs # we compare ket1 and ket3 pval, adj, _ = nbs.nbs_bct(trt1Reshape, trt2Reshape, thresh=2.5, tail='both',k=500, paired=True, verbose = True) # one network is different #%% compare sad to trauma 1 sad1Reshape = np.moveaxis(np.array(sad_corr_subs),0,-1) pvalSadTr, adjSadTr, _ = nbs.nbs_bct(trt1Reshape, sad1Reshape, thresh=2.5, tail='both',k=500, paired=True, verbose = True) # sig difference between those two networks. # now contrast between the mean matrices of each condition contMat = np.mean(trauma_1st_ses, axis = 0) - np.mean(sad_corr_subs, axis = 0) # then multiply by the adjacency matrix created by NBS. adjCor = contMat * adjSadTr np.max(adjCor) # now we can differentiate to two positive and negative matrices pos_cor = np.array(adjCor) pos_cor[pos_cor<=0] = 0 # zero for everything lower than zero np.max(pos_cor)
# using the condition labels to seperate midazolam and ketamine
ketamine_mat = []
midazolam_mat = []
for i,x in enumerate(condition_label):
print(i)
print(x)
if x==1:
# ketamine
ketamine_mat.append(mat_2[i])
else:
midazolam_mat.append(mat_2[i])
#%% now reshape and NBS
ketamine = np.moveaxis(np.array(ketamine_mat),0,-1)
midazolam = np.moveaxis(np.array(midazolam_mat),0,-1)
#%%
from bct import nbs
# we compare ket1 and ket3
pval, adj, _ = nbs.nbs_bct(ketamine, midazolam, thresh=2.5, tail='both',k=500, paired=False,
verbose = True)
# no difference in RS across groups
# %%
nilearn.plotting.plot_matrix(mat_2[0], labels=np.array(labels.Yeo_networks7) ,
colorbar=True)
#%%
np.array(labels.Difumo_names)