Figure 8 - Functional imaging

This notebook loads in the data corresponding to figure 8, both the low and high magnification data sets.

We then compute:

The mean intenstity image of the low magnification 8A)
The crosscorrelation image with region 1, 8B)
The mean intenstity image of the high magnification 8C) with rois
The example traces and roi correlation histograms

Setup plotting



In [30]:

    
from showit import image
%matplotlib inline



In [31]:

    
import seaborn as sns
sns.set_context('paper', font_scale=2.0)
sns.set_style('ticks')

Load in low magnigication data



In [3]:

    
from thunder.images import frombinary

path = 's3://neuro-datasets/svoboda-lab/2pRAM-paper/00'

# we only load in the first 100 out of 2628 frames as the data is large
data = frombinary(path + '/images', stop = 100)



In [4]:

    
data









    Out[4]:





Images
mode: local
dtype: int16
shape: (100, 500, 468)

Compute and inspect the mean image



In [5]:

    
mean = data.mean().toarray()



In [6]:

    
image(mean, clim=(0, 2500), size = 12);

Average activity in subregions



In [7]:

    
from json import loads
from regional import many
from requests import get

path = 'https://s3.amazonaws.com/neuro-datasets/svoboda-lab/2pRAM-paper/00'

r = get(path + '/sources/sources.json').content
sources = loads(r)

regions = many([x['coordinates'] for x in sources])



In [8]:

    
img = regions.mask(base=(mean/2500).clip(0,1), stroke='orange', fill = None)
image(img, size = 12);



In [9]:

    
from extraction.model import ExtractionModel

em = ExtractionModel(regions)
traces = em.transform(data)



In [10]:

    
from numpy import array
time = array([float(x)/4.38 for x in range(traces.shape[2])])



In [11]:

    
from matplotlib.pyplot import plot, xlim, xlabel, ylabel 

plot(time, traces[0, 0].values.T.squeeze());
xlabel('time (s)', fontsize=14);
xlim([0, 25])
ylabel('F (a.u.)', fontsize=14);

Make correlation map



In [12]:

    
records = data.uniform_filter(5).toseries()



In [13]:

    
records









    Out[13]:





Series
mode: local
dtype: int16
shape: (500, 468, 100)



In [14]:

    
corr = records.correlate(traces[0, 0].values[0]).toarray()



In [15]:

    
image(corr, clim = (0, 0.7), size = 12);

Load in high magnigication data



In [16]:

    
from thunder.images import frombinary

path = 's3://neuro-datasets/svoboda-lab/2pRAM-paper/01'

# we only load in the first 100 out of 6000 frames as the data is large
data = frombinary(path + '/images' , stop = 100)



In [17]:

    
data









    Out[17]:





Images
mode: local
dtype: int16
shape: (100, 4, 512, 512)

Perform motion correction



In [18]:

    
from registration import CrossCorr

# use first 60 frames as a reference
reference = data[:60].mean().toarray()

# perform registration on each plane independently
algorithm = CrossCorr(axis=0)
model = algorithm.fit(data, reference)
registered = model.transform(data)



In [19]:

    
mean = registered.mean().toarray()



In [20]:

    
image(mean[3], clim=(0, 2500), size = 12);

Display sources



In [21]:

    
from json import loads
from regional import many
from requests import get

path = 'https://s3.amazonaws.com/neuro-datasets/svoboda-lab/2pRAM-paper/01'


regions = []
for i in range(4):
    r = get(path + '/sources/sources-%d.json' % i).content
    sources = loads(r)
    regions.append(many([x['coordinates'] for x in sources]))



In [22]:

    
plane = 3
img = regions[plane].mask(base=(mean[plane]/2500).clip(0,1), stroke='orange', fill = None)
image(img, size = 12);

Compute and inspect neuron activity traces



In [23]:

    
from extraction.model import ExtractionModel

traces = []
for plane in range(4):
    em = ExtractionModel(regions[plane])
    trace = em.transform(registered[:,plane,:,:].squeeze()).normalize('window', window=500)
    traces.append(trace)



In [24]:

    
from numpy import array
time = array([float(x)/9.66 for x in range(traces[0].shape[2])])



In [25]:

    
from matplotlib.pyplot import plot, xlim, xlabel, ylabel 

plane = 0
neuron = 20

plot(time, traces[plane][0, neuron].values.T.squeeze());
xlabel('time (s)', fontsize=14);
xlim([0, 10])
ylabel('DFF', fontsize=14);

Extract temporal correlations



In [26]:

    
# compute maximum activity in region 1
avg = traces[0].max()



In [27]:

    
from matplotlib.pyplot import plot, xlim, xlabel, ylabel 

plot(time, avg.values.T.squeeze());
xlabel('time (s)', fontsize = 14);
xlim([0, 10])
ylabel('DFF', fontsize = 14);



In [28]:

    
# compute correlation with timeseries of each individual neuron
corr = [t.correlate(avg).values for t in traces]

Create histograms of correlation coefficients



In [29]:

    
from numpy import linspace
from matplotlib.pyplot import hist, xlim, xlabel, ylabel 

plane = 2
hist(corr[plane][0],bins=linspace(-.525,.525,22), rwidth=0.9, edgecolor='none', color = [0.7,0.7,0.7]);
xlim([-.35, .35])
xlabel('Correlation', fontsize = 14);
ylabel('Neurons', fontsize = 14);