import pickle, json
import matplotlib.pyplot as plt
from netpyne import specs,sim
import numpy as np
import scipy
import os
import umap
import random
Trials = ['0_0','0_1','0_2','0_3','0_4']
Ndata = 1 # One condition
reading_folder = ['Data_Control'] #, 'Data_MotorThalamusInactivation', 'Data_NoradrenalineBlock']
pick_label_pre = ['v56_batch23_0_0_'] #, 'v56_batch20_0_1_0_', 'v56_batch22_']
pick_label = []
for label in pick_label_pre:
for trial in Trials: pick_label.append(label+trial)
Twindow = 25
WindowOverlap = 0
Tstart = 3000
Ton = 5000
Tend = 7000
## DATA PREPROCESSING
data = {}
spkt = {}
spkid = {}
NcellsSample = 1000
## UMAP
n_neighbors = 100
min_dist = 0.95
metric = 'euclidean'
###############################################################################
# Structuring analysis
if isinstance(Twindow,int): Twindow = [Twindow]
for label in pick_label:
if label.startswith('v56_batch23'):
folder = os.path.join(os.getcwd(),reading_folder[0])
filename = label+'.pkl'
filename = os.path.join(folder,filename)
print(('Loading file %s ... ' % (filename)))
with open(filename, 'rb') as fileObj:
data[label] = pickle.load(fileObj, encoding='latin1')
if label.startswith('v56_batch20'):
folder = os.path.join(os.getcwd(),reading_folder[1])
filename = label+'.json'
filename = os.path.join(folder,filename)
print(('Loading file %s ... ' % (filename)))
with open(filename, 'rb') as fileObj:
data[label] = json.load(fileObj, encoding='latin1')
if label.startswith('v56_batch22'):
folder = os.path.join(os.getcwd(),reading_folder[2])
filename = label+'.json'
filename = os.path.join(folder,filename)
print(('Loading file %s ... ' % (filename)))
with open(filename, 'rb') as fileObj:
data[label] = json.load(fileObj, encoding='latin1')
spkt[label] = np.array(data[label]['simData']['spkt'])
spkid[label] = np.array(data[label]['simData']['spkid'])
###############################################################################
# Creating the populations
L5Bmin=0.47
L5Bmax=0.8
L5Bmid = L5Bmin + (L5Bmax-L5Bmin)/2
pops_labels = ['IT2','SOM2','PV2','IT4','IT5A','SOM5A','PV5A','IT5B','upperPT5B','lowerPT5B','SOM5B','PV5B','IT6','CT6','SOM6','PV6']
Npops = len(pops_labels)
# ReCreate the network according to batch23
if pick_label[0].startswith('v56_batch23'):
cfg = specs.SimConfig(data[pick_label[0]]['simConfig'])
cfg.createNEURONObj = False
sim.initialize() # create network object and set cfg and net params
sim.loadAll('', data=data[pick_label[0]], instantiate=False)
sim.setSimCfg(cfg)
sim.net.createPops()
sim.net.createCells()
pops = specs.ODict({'IT2' :[c.gid for c in sim.net.cells if c.tags['pop']=='IT2'],
'SOM2' :[c.gid for c in sim.net.cells if c.tags['pop']=='SOM2'],
'PV2' :[c.gid for c in sim.net.cells if c.tags['pop']=='PV2'],
'IT4' :[c.gid for c in sim.net.cells if c.tags['pop']=='IT4'],
'IT5A' :[c.gid for c in sim.net.cells if c.tags['pop']=='IT5A'],
'SOM5A' :[c.gid for c in sim.net.cells if c.tags['pop']=='SOM5A'],
'PV5A' :[c.gid for c in sim.net.cells if c.tags['pop']=='PV5A'],
'IT5B' :[c.gid for c in sim.net.cells if c.tags['pop']=='IT5B'],
'upperPT5B':[c.gid for c in sim.net.cells if L5Bmin <= c.tags['ynorm'] <= L5Bmid and c.tags['pop']=='PT5B'],
'lowerPT5B':[c.gid for c in sim.net.cells if L5Bmid <= c.tags['ynorm'] <= L5Bmax and c.tags['pop']=='PT5B'],
'SOM5B' :[c.gid for c in sim.net.cells if c.tags['pop']=='SOM5B'],
'PV5B' :[c.gid for c in sim.net.cells if c.tags['pop']=='PV5B'],
'IT6' :[c.gid for c in sim.net.cells if c.tags['pop']=='IT6'],
'CT6' :[c.gid for c in sim.net.cells if c.tags['pop']=='CT6'],
'SOM6' :[c.gid for c in sim.net.cells if c.tags['pop']=='SOM6'],
'PV6' :[c.gid for c in sim.net.cells if c.tags['pop']== 'PV6'],
# External inputs
'TPO' :[c.gid for c in sim.net.cells if c.tags['pop']== 'TPO'],
'TVL' :[c.gid for c in sim.net.cells if c.tags['pop']== 'TVL'],
'S1' :[c.gid for c in sim.net.cells if c.tags['pop']== 'S1'],
'S2' :[c.gid for c in sim.net.cells if c.tags['pop']== 'S2'],
'cM1' :[c.gid for c in sim.net.cells if c.tags['pop']== 'cM1'],
'M2' :[c.gid for c in sim.net.cells if c.tags['pop']== 'M2'],
'OC' :[c.gid for c in sim.net.cells if c.tags['pop']== 'OC']
})
NcellsTotal = sum([len(pops[pop]) for pop in pops])
cells_pop = [0]*NcellsTotal
for pop in pops:
for nn in range(len(pops[pop])):
cells_pop[pops[pop][nn]] = pop
Tduration = data[pick_label[0]]['simConfig']['duration']
Ncells = sum(len(pops[pop]) for pop in pops_labels) # excluding stims
###############################################################################
# Selected neurons
validPops = pops_labels
Id_validCells = [id for id in range(NcellsTotal) if cells_pop[id] in validPops]
activityVector = {}
for Tw in Twindow:
print(Tw)
t_stamps = np.linspace(Tw/2.0,Tduration-Tw/2.0,int((Tduration-Tw)/((1.0-WindowOverlap)*Tw))+1,endpoint=True)
t_limits = [(t_stamps[nt]-Tw/2.0,t_stamps[nt]+Tw/2.0) for nt in range(len(t_stamps))]
index_start = [Tstart<=t_stamps[nt] for nt in range(len(t_stamps))].index(True)
index_end = [Tend<=t_stamps[nt] for nt in range(len(t_stamps))].index(True)
activityVector[str(Tw)] = {}
for label in pick_label:
activityVector[str(Tw)][label] = np.zeros((Ncells,len(t_stamps)))
for n in range(len(spkt[label])):
neuron = int(spkid[label][n])
if cells_pop[neuron] in pops_labels: # cells and not stimuli
nt = [t_limits[nt][0] <= spkt[label][n] < t_limits[nt][1] for nt in range(len(t_limits))].index(True)
activityVector[str(Tw)][label][neuron][nt] = activityVector[str(Tw)][label][neuron][nt] + 1000.0/Tw
if label == pick_label[0]:
Matrix = activityVector[str(Tw)][label][Id_validCells,index_start:index_end]
else:
Matrix = np.concatenate((Matrix,activityVector[str(Tw)][label][Id_validCells,index_start:index_end]),axis=1)
mean_activity = np.sum(Matrix,1)*(Tw/1000.0) / ((Tend-Tstart)/1000.0) # number of spikes in [Tstart,Tend] / T
Valid_Ids = [n for n in range(len(mean_activity)) if mean_activity[n]!=0]
Ids = random.sample(Valid_Ids, NcellsSample)
Ids.sort()
x = np.transpose(Matrix[Ids,:])
for n_components in [3]: #[2,3]:
umap_reduction = umap.UMAP(n_neighbors=n_neighbors,min_dist=min_dist,n_components=n_components,metric=metric).fit(x)
umap_representation = umap_reduction.transform(x)
umap_representation_back = umap_reduction.inverse_transform(umap_representation)
# Plot as in the paper (not exactly the same figure because sampled cells are different)
if n_components==3 and Tw==25:
time = t_stamps[index_start:index_end]
index_on = [Ton<=time[nt] for nt in range(len(time))].index(True)
index_end = len(time)
orig_data = {}
umap_data = {}
umap_back_data = {}
for index,label in enumerate(pick_label):
orig_data[label] = x[index*len(time):(index+1)*len(time),:]
umap_data[label] = umap_representation[index*len(time):(index+1)*len(time),:]
umap_back_data[label] = umap_representation_back[index*len(time):(index+1)*len(time),:]
# Plot data
color1 = (0.175, 0.35, 0.66, 1)
color2 = (0.90, 0.21, 0.16, 1)
lab = pick_label_pre[0]
trial = '0_0' # only first trial
label = str(lab) + str(trial)
import csaps
theta = np.arange(len(umap_data[label]))
data_traj = np.transpose(umap_data[label])
sp_theta = csaps.MultivariateCubicSmoothingSpline(data_traj, theta, smooth=1.0)
theta_i = np.linspace(0, len(umap_data[label])-1, 10000)
data_i = np.transpose(sp_theta(theta_i))
for az in np.linspace(0,360,37):
fig = plt.figure(figsize=(20, 16), dpi=300)
ax = fig.gca(projection='3d')
ax.azim = az
ax.dist = 10
ax.elev = 25
ax.scatter3D(umap_data[label][0:index_on,0], umap_data[label][0:index_on,1], umap_data[label][0:index_on,2], color=color1, s=25, alpha = 1.0)
ax.scatter3D(umap_data[label][index_on:index_end,0], umap_data[label][index_on:index_end,1], umap_data[label][index_on:index_end,2], color=color2, s=25, alpha = 1.0)
ax.plot3D(data_i[0:int(len(data_i)/2),0], data_i[0:int(len(data_i)/2),1], data_i[0:int(len(data_i)/2),2], color=color1, linewidth=1.0)
ax.plot3D(data_i[int(len(data_i)/2):len(data_i),0], data_i[int(len(data_i)/2):len(data_i),1], data_i[int(len(data_i)/2):len(data_i),2], color=color2, linewidth=1.0)
startpoint = ax.scatter3D(umap_data[label][0,0], umap_data[label][0,1], umap_data[label][0,2], color=color1, s=150, alpha = 1.0, label='Start')
transitionpoint = ax.scatter3D(umap_data[label][index_on,0], umap_data[label][index_on,1], umap_data[label][index_on,2], color=color2, edgecolor = 'k', linewidth=2, s=150, alpha = 1.0, label='Transition')
endpoint = ax.scatter3D(umap_data[label][index_end-1,0], umap_data[label][index_end-1,1], umap_data[label][index_end-1,2], color=color2, s=150, alpha = 1.0, label='End')
ax.set_top_view()
ax.set_xlabel('$C_1$', fontsize=40, labelpad=35)
ax.set_xticklabels([])
ax.set_ylabel('$C_2$', fontsize=40, labelpad=35)
ax.set_yticklabels([])
ax.set_zlabel('$C_3$', fontsize=40, labelpad=35)
ax.set_zticklabels([])
ax.w_xaxis.set_pane_color((1.0, 1.0, 1.0, 0.0))
ax.w_yaxis.set_pane_color((1.0, 1.0, 1.0, 0.0))
ax.w_zaxis.set_pane_color((1.0, 1.0, 1.0, 0.0))
plt.savefig('ReducedDim-'+str(az)+'.png')
plt.close(fig)
# Reconstruction
fig, (ax1, ax2) = plt.subplots(1,2, dpi=300)
minSpikeCount = -5
maxSpikeCount = 75
(pearsonCorr,pvalue) = scipy.stats.pearsonr(orig_data[label].flatten(), umap_back_data[label].flatten())
fig.set_figheight(7.5)
fig.set_figwidth(15)
ax1.set_title('Original activity data', fontsize=40, pad=25)
im1 = ax1.imshow(np.transpose(orig_data[label][:,:]),cmap='viridis', aspect='auto')
cbar1 = plt.colorbar(im1,ax=ax1)
cbar1.ax.tick_params(labelsize=35)
ax1.set_xticks([-0.5,len(time)/2 - 0.5,len(time) - 0.5])
ax1.set_xticklabels([0 , 2, 4])
ax1.set_yticks([0,250,500,750,1000])
ax1.tick_params(labelsize=35)
ax1.set_xlabel('Time (s)', fontsize=40)
ax1.set_ylabel('Neuron ID', fontsize=40)
im1.set_clim(minSpikeCount, maxSpikeCount)
ax1.axvline(len(time)/2,color='white',dashes=[8, 4],linewidth=4)
ax2.set_title(r'Reconstruction ($\rho$ = %.3f)' % (pearsonCorr), fontsize=40, pad=25)
im2 = ax2.imshow(np.transpose(umap_back_data[label][:,:]),cmap='viridis', aspect='auto')
cbar2 = plt.colorbar(im2,ax=ax2)
cbar2.ax.tick_params(labelsize=35)
ax2.set_xticks([-0.5,len(time)/2 - 0.5,len(time) - 0.5])
ax2.set_xticklabels([0 , 2, 4])
ax2.set_yticks([0,250,500,750,1000])
ax2.tick_params(labelsize=35)
ax2.set_xlabel('Time (s)', fontsize=40)
cbar2.set_label('Rate (Hz)', fontsize=40)
im2.set_clim(minSpikeCount, maxSpikeCount)
ax2.axvline(len(time)/2,color='white',dashes=[8, 4],linewidth=4)
plt.tight_layout()
plt.savefig('ReconstructionMap.png')
plt.close(fig)