[Comp-neuro] Announcement of a new paper on correlations

ramana dodla ramana.dodla at gmail.com
Wed Feb 24 00:22:11 CET 2010

I'm quite excited to post here the abstract of a new paper, because talking
correlations would bring to life so much discussion, and we claimed in the
paper that the method described there (a ``phase function'') produces better
results than a regular correlation function applied to spike times. Of
particular mention is the usefulness of this method when the number of spike
times is very few, say less than 10. For 10 spike times, for example, an
auto-correlation function will not be very useful, but a phase function
be able to show a clear oscillatory profile without resorting to any

Ready to use codes in Matlab/Octave and Mathematica are available as
supplementary material for the paper, and also in the link given below.
better, a JavaScript page is online (link below) to quickly try out the
efficacy of
this method. Your feedback is most welcome. Thank you....  ramana dodla

PS: Please see the cover-art of the journal for bigger plots of phase
applied to ISIs as few as 3, and as many as more than 600.

 TITLE: A Phase Function to Quantify Serial Dependence between Discrete
 AUTHORS: Ramana Dodla and Charles J. Wilson
 JOURNAL: Biophysical Journal 98:L05-L07, 2010
 URL: http://dx.doi.org/10.1016/j.bpj.2009.11.003

Auto- and cross-correlation methods, when applied to discrete events, can
determine periodicity and correlation times within and between event train
sequences. However, if the number of available events for analysis is too
the correlation techniques yield ambiguous and insufficient results. Here we
report a technique based on measurements of phases of event times that could
detect the periodicity even among very few discrete data points. The results
are demonstrated on in vitro neuronal spike time data, and are found to be
highly contrasting when compared with the correlation techniques. The
could become invaluable, for example, for treating in vivo spike time
that often last very short duration, or for determining short timescales in
discrete biophysical experimental data.

A method based on relative phases between discrete time events is used to
capture the serial correlations in an experimental spike train. The figure
depicts phase correlations of a spike train with itself as a function of
lag by including an increasing number of time events. The top five traces
autophase function curves for a spike train with as few as 3, 4, 5, 6, and 7
time events, respectively. The bottom three traces use 14, 68, and 673 time
events, respectively. For a large number of events, the phase function
a profile equivalent to a corresponding correlation function. But for a very
few number of events, it can show a measurable profile where a correlation
function may fail to reliably detect one.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.neuroinf.org/pipermail/comp-neuro/attachments/20100223/fb87eff2/attachment.html

More information about the Comp-neuro mailing list