def read_ep_nodal(ep_subj, filename):
"""
This function creates an empty dataframe, builds the brain graphs,
calculates the network features and writes them in the dataframe.
Created by: Loukas Serafeim, Nov 2017
Args:
ep_subj: A list with subject numbers. e.g. for 3 subjects use ep_subj= [1, 2, 3]
filename: This is the filename of the subjects. e.g. filename = 'ep_subj_{}'
Returns:
3 network features per node: deg, cc, ne for the EP group and concatinate them in one dataframe
"""
counter = 0
for s in ep_subj:
X = pd.read_csv(filename.format(s), header=None)
X = X.values
np.fill_diagonal(X, 0)
X = X.tolist()
g = Graph.Weighted_Adjacency(X,
mode=ADJ_UNDIRECTED,
attr="weight",
loops=False)
N = g.vcount()
E = g.ecount()
g.es["label"] = g.es["weight"]
lst = range(N)
g.vs["name"] = lst
d = g.strength(loops=False, weights=g.es["weight"])
cc = g.transitivity_local_undirected(vertices=None,
mode="zero",
weights=g.es["weight"])
ne = nodal_eff(g)
d = np.array(d)
cc = np.array(cc)
if counter == 0:
dt2 = pd.DataFrame(columns=['Deg'] * d.shape[0] +
['Cc'] * d.shape[0] + ['Ne'] * d.shape[0] +
['ID'])
all_together = np.concatenate((d, cc, ne), axis=0)
all_together = np.append(all_together, 2)
dt2.loc[s] = list(all_together)
counter += 1
return dt2
def decompose_lobes(grouping, subj_numbers, filename, verbose=0):
c = [
item for sublist in [[
'deg_{}_lob'.format(s), 'cc_{}_lob'.format(s), 'ne_{}_lob'.format(
s)
] for s in range(len(np.unique(grouping).tolist()))]
for item in sublist
]
c.append('id')
dt_lobe = pd.DataFrame(columns=c)
for j in subj_numbers:
Xc = pd.read_csv(filename.format(j), header=None)
Xc = Xc.values
np.fill_diagonal(Xc, 0)
Xc = Xc.tolist()
gc = Graph.Weighted_Adjacency(Xc,
mode=ADJ_UNDIRECTED,
attr="weight",
loops=False)
Nc = gc.vcount()
Ec = gc.ecount()
gc.es["label"] = gc.es["weight"]
lst_c = range(Nc)
gc.vs["name"] = lst_c
lists = []
for i in range(len(np.unique(grouping))):
tmp_cluster = np.where(grouping == i)[0].tolist()
#print(tmp_cluster)
sub = gc.induced_subgraph(tmp_cluster)
av_d = average(sub.strength(loops=False, weights=sub.es["weight"]))
av_cc = average(
sub.transitivity_local_undirected(vertices=None,
mode="zero",
weights=sub.es["weight"]))
av_ne = average(nodal_eff(sub))
lists.extend([av_d, av_cc, av_ne])
if filename == 'ctrl_FA_{}.csv':
lists.append(1)
elif filename == 'ep_FA_{}.csv':
lists.append(2)
elif filename == 'iugr_FA_{}.csv':
lists.append(3)
elif filename == 'ctrl_VAV_{}.csv':
lists.append(4)
dt_lobe.loc[j] = lists
return dt_lobe, grouping
def read_ep(ep_subj, filename):
"""
This function creates an empty dataframe, builds the brain graphs,
calculates the network features and writes them in the dataframe.
Created by: Loukas Serafeim, Nov 2017
Args:
ep_subj: A list with subject numbers. e.g. for 3 subjects use ep_subj= [1, 2, 3]
filename: This is the filename of the subjects. e.g. filename = 'ep_subj_{}'
Returns:
3 Global network features: deg, cc, ne for the EP group and concatinate them in one dataframe
"""
dt2 = pd.DataFrame(columns=['Deg', 'Cc', 'Ne', 'ID'])
for s in ep_subj:
X = pd.read_csv(filename.format(s), header=None)
X = X.values
np.fill_diagonal(X, 0)
X = X.tolist()
g = Graph.Weighted_Adjacency(X,
mode=ADJ_UNDIRECTED,
attr="weight",
loops=False)
N = g.vcount()
E = g.ecount()
g.es["label"] = g.es["weight"]
lst = range(N)
g.vs["name"] = lst
d = g.strength(loops=False, weights=g.es["weight"])
av_d = average(d)
cc = g.transitivity_local_undirected(vertices=None,
mode="zero",
weights=g.es["weight"])
av_cc = average(cc)
ne = nodal_eff(g)
av_ne = average(ne)
dt2.loc[s] = [av_d, av_cc, av_ne, 2]
return dt2
def decompose_iugr_lev(X, iugr_subj, filename, verbose=0):
"""
This function decomposes the IUGR subjects using the leading eigenvector algorithm
based on the prior information.
Created by: Loukas Serafeim, Nov 2017
Args:
X: The MEAN ctrl network
iugr_subj: A list with the IUGR subject numbers. e.g. for 3 subjects use iugr_subj= [1, 2, 3]
filename: This is the filename of the IUGR subjects. e.g. filename = 'iugr_subj_{}'
Returns:
The average degree, clustering coefficient and nodal efficiency of each community
(3*comms features)
"""
# c =['deg', 'cc','ne']*6
# c.append('id')
# dt_iugr_lev = pd.DataFrame(columns = c)
X = X.values
np.fill_diagonal(X, 0)
X = X.tolist()
g = Graph.Weighted_Adjacency(X,
mode=ADJ_UNDIRECTED,
attr="weight",
loops=False)
N = g.vcount()
E = g.ecount()
g.es["label"] = g.es["weight"]
g.vs["name"] = range(N)
comms = g.community_leading_eigenvector(weights=g.es["weight"])
clusters = comms
if verbose:
print(
"\nafter calculating the average control matrix the comms are: ")
print(clusters)
#c =['deg', 'cc','ne'] * len(clusters)
c = [
item for sublist in
[['deg_{}'.format(s), 'cc_{}'.format(s), 'ne_{}'.format(s)]
for s in range(len(clusters))] for item in sublist
]
c.append('id')
dt_iugr_lev = pd.DataFrame(columns=c)
for j in iugr_subj:
Xc = pd.read_csv(filename.format(j), header=None)
Xc = Xc.values
np.fill_diagonal(Xc, 0)
Xc = Xc.tolist()
gc = Graph.Weighted_Adjacency(Xc,
mode=ADJ_UNDIRECTED,
attr="weight",
loops=False)
Nc = gc.vcount()
Ec = gc.ecount()
gc.es["label"] = gc.es["weight"]
lst_c = range(Nc)
gc.vs["name"] = lst_c
lists = []
for i in range(len(clusters)):
sub = gc.induced_subgraph(clusters[i])
av_d = average(sub.strength(loops=False, weights=sub.es["weight"]))
av_cc = average(
sub.transitivity_local_undirected(vertices=None,
mode="zero",
weights=sub.es["weight"]))
av_ne = average(nodal_eff(sub))
lists.extend([av_d, av_cc, av_ne])
lists.append(3)
dt_iugr_lev.loc[j] = lists
return dt_iugr_lev, clusters
