Biomedical Engineering Reference
In-Depth Information
observed and shuffled data sets. This procedure produced a sin-
gle metric for the analyses of significant correlations. For shuffled
data, the ratio of peak size of 1.1 corresponded to a p
0. 05.
We interpreted ratios larger than this to be significant. A total
of 35 pairs had significantly larger peaks than could be expected
by chance. However, few of these pairs had strong correlations
(mean peak to noise ratio: 1.23) (
Fig. 7.4E
).
When using the cross-correlations, several caveats must be
considered. The first is that cross-correlation assumes station-
arity of neurons. Under typical experimental conditions, many
non-stationarities are present (anesthesia state, behavioral moti-
vation, unit waveform drift, satiety states). Secondly, changes in
rate can lead to spurious rate correlations - for this reason, a
'shift-predictor', i.e., the spike train from one neuron is shifted
by the distance of one trial in time, is often subtracted from the
cross-correlations to differentiate true interactions from simply
rate covariations. Thirdly, in cross-correlative analyses, a sufficient
number of spikes in both spike trains are required for statistical
inference (Gerstein 1985). Fourthly, as discussed above, cross-
correlation inferences rely on clear, artifact-free unit isolation.
Fifth, the time-scale of neuronal interactions may influence the
cross-correlation
(26)
.
Cross-correlation can also be used to look at long timescale
interactions, i.e., on the order of several milliseconds or tens
of seconds. Such long timescale interactions do not represent
synaptic interactions; rather, they represent an increase between
the firing rates of two neurons. While interesting, other tech-
niques (such as rate correlations) may be better suited to such
interferences.
These results illustrate how cross-correlation can be used
to make inferences about connectivity in multi-electrode data.
While this approach is perhaps the most common to studying
functional interactions, the cross-correlation does not incorporate
how correlations change as a function of behavior. To apprehend
the dynamics of correlation, we use the joint-peristimulus time
histogram.
<
6.
Joint-Peristimulus
Time Histogram
Since behavior provides a key window into neural responses (via
perievent perievent or peristimulus peristimulus histograms), it
is of great interest to know how relationships between neu-
rons vary as a function of stimuli or events. Aertsen and
colleagues
(27)
created the joint-peristimulus time histogram
