# regularity of spike series

Glen M. Sizemore gmsizemore2 at yahoo.com
Sat Jan 24 10:30:13 EST 2004

```Mat: Autocorrelograms can reveal simple regularities in spike trains, e.g.
if many spikes are followed by another one at 30s then you'll get a
that what I need is a measure of higher-order patterns, though I'm not
sure even if I find one how to interpret that since it might be highly
non-intuitive.

GS: Back up. What does the inter-event interval distribution look like? Make
sure you plot the data using the right "bin size;" too large a bin and
multiple modes can be obscured. You don't want the bin size to be too small
either. The distribution is the first place to look for "simple regularities
in spike trains." I don't think it is necessarily true that "...if many
spikes are followed by another one at 30s then you'll get a bump in the
autocorrelogram," is it? If we construct a time-series by randomly drawing
from a normal distribution centered at 30 s, we would not see a reliable
bump anywhere when we plotted an autocorrelogram constructed from the
time-series, as there would be no sequential dependencies - the duration of
an event at n has no bearing on the duration at n+x, no? Am I missing
something? Plot the distribution of inter-event intervals, then calculate
the inter-event interval per opportunity (IEI/op). That is, plot how many
IEIs occurred in a particular bin and divide it by the number of IEIs in
that bin plus all those in the bins greater than it. Plot this number as a
function of ordinal position of the bin (and, of course, the ordinal
position is related to time as that is the dimension of the bin). This
yields a conditional probability at each discrete "point" in time
(obviously, the accuracy depends on the size of the bin). It expresses the
probability that an event will follow the previous event by an amount of
time defined by the ordinal position of the bin. This tells you something
about the likelihood that an event will occur at any point following another
event. If you don't get any reliable auto-correlation at any lag, it means
that the distribution specifies everything about the temporal properties of
the train of events that is possible under those conditions. No?

"mat" <mats_trash at hotmail.com> wrote in message
> y.k.y at lycos.com (yan king yin) wrote in message
> > Spike trains obtained from a single neuron is the
> > result of dendritic integration of ~1000 to 100,000
> > of other neurons' inputs. What kind of information can
> > autocorrelation of a few spike trains reveal? Probably
> > very little...
> >
> > I guess your research will be more fruitful if you
> > focus more on single neuron information processing.
> >
>  These are not single neuron spikes they are field potentials.  I'm
> using an in vitro model of epileptic activity by perfusing magnesium
> free artificial CSF.  What I would like to establish is whether there
> is any degree of regularity to these trains of depolarizations
> (~6-10/min) and see if this is altered when I co-perfuse drugs across
> the cortex.
>
> Autocorrelograms can reveal simple regularities in spike trains, e.g.
> if many spikes are followed by another one at 30s then you'll get a
> bump in the autocorrelogram.  But since I've not found anything like
> that what I need is a measure of higher-order patterns, though I'm not
> sure even if I find one how to interpret that since it might be highly
> non-intuitive.

```