Methods for Studying Functional Interactions Among Neuronal Populations - Dynamic Brain Imaging - page 146

data1 = rand(70, 10); % random peri-event matrix: 70
trials, 10 bins
data2 = rand(70, 10); % random peri-event matrix: 70
trials, 10 bins
fr_data1 = mean(data1'); % sums along rows
fr_data2 = mean(data2'); % sums along rows
scatter(fr_data1, fr_data2); corrcoef(fr_data1,
fr_data2)

We applied trial-by-trial rate correlations to our sample
dmPFC and motor cortex recordings over 235 correct trials;
this number yielded similar power as our rate correlation anal-
ysis (above); over each trial, animals performed a stereotyped
series of actions. In trial-shuffled data, we found that correla-
tions greater than 0. 05. Of 210

interactions, 56 pairs (27%) had significant trial-by-trial rate cor-
relations (69% of dmPFC pairs, 16% of motor cortex pairs, and
43% of dmPFC-motor pairs). Neurons exhibited strong trial-by-
trial covariations. For instance, two neurons could be strongly
correlated ( Fig. 7.7A ) or anti-correlated ( Fig. 7.7B ) over the
session; these are the same neurons with rate correlations. Most
neurons did not have strong correlations ( Fig. 7.7C) . Note that
while motor cortex neurons had strong rate correlations, they
tended not have strong trial-by-trial interactions. These meth-
ods, then, are able to separate resolve patterns of functional
correlations.

| 0. 18 | corresponded to p <

9. Predictive
Interactions

Thus far, we have discussed functional interactions in three
domains - patterns of activity (PCA), timing of spikes (cross-
correlation / JPSTH) and fluctuations in firing rate (rate and
trial-by-trial covariations). Though these approaches are power-
ful in examining relationship between neurons, none of these
approaches examines how neurons predict behavior. If neurons
predict behavior in similar ways, then they may have functional
interactions that are due to behavior.
To investigate relationships in how neurons predict behav-
ior, we used methods from the field of machine learning ,or
statistical pattern recognition , to quantify information from spike
trains about behavior on a trial-by-trial basis (see Refs. (5) for
review of methods for spike train analysis and Refs. (32-36) for
a review of the statistical issues). By comparing the predictions
of behavior between two neurons, we can investigate whether
neurons predict behavior in the same way (i.e., predict the same
behavior on the same trials), or in different ways (i.e., predict