Biomedical Engineering Reference
In-Depth Information
The above steps are simple and can be rapidly computed:
t1 = sort(rand(1, 200)
*
10); % random
timestamps,10s @ 20 Hz
t2 = sort(rand(1, 50)
*
10); % random timestamps,10s @ 5
Hz
edges = [0.25:0.25:10]; % edges for 50 ms bins
rate_t1 = histc(t1, edges); % bins for t1
rate_t2 = histc(t2, edges); % bins for t2
scatter(rate_t1, rate_t2); corrcoef(rate_t1,
rate_t2)
In the above code, two spike trains are generated, binned,
and then correlated. This is a rapid way to look for relationships
between the firing rates of two neurons and also can be useful in
addressing non-stationarities that may threaten cross-correlation
or JPSTH analyses.
We applied simple rate correlations to dmPFC and motor
cortex recordings. We binned spike trains over 10 s over the
39 minutes animals were behaving, resulting in binned firing
rates over a total of 235 bins. Note that over this epoch, con-
siderable behavioral fluctuations could occur; however, for the
present, we are only interested in relationships between neurons,
and ignore all other parameters. In trial-shuffled data, correlations
greater than
0. 05. Of 210 interac-
tions, 109 pairs (52%) had significant rate correlations (67% of
dmPFC pairs, 52% of motor cortex pairs, and 45% of dmPFC-
motor pairs). Neurons could exhibit strong covariations over the
session. For instance, two neurons could be strongly correlated
(
Fig. 7.6B
) or anti-correlated (
Fig. 7.6C
) over the session.
Most neurons did not have strong correlations (
Fig. 7.6D
).
This result suggests that overall correlations were not governed
by gross non-stationarities over behavioral sessions (i.e., degra-
dation of unit isolation, artifacts in recording, decline in moti-
vation); rather, the structure of correlations between specific
pairs suggested that pairs of neurons exhibited strong functional
interactions.
Perhaps the largest difficulty with rate correlations is that the
animal behavior is uncontrolled. Correlations may be driven by
behavioral variability, similar functional properties across neurons
(compare with PCA), or even task-irrelevant behavior, such as
grooming or movement artifacts. For these reasons, it is some-
what difficult to interpret rate correlations, especially in the
absence of well-controlled operant behavior. In using rate corre-
lations, all previous caveats apply (unit isolation, sufficient spikes).
Furthermore, the number of bins should be carefully chosen. Too
few bins lead to inflated correlation coefficients. Too many bins
may destroy temporal structure. For this reason, any observed
correlation must be explored at a variety of bin sizes.
|
0. 14
|
corresponded to
p
<
