Analysis of Recorded Spike Trains

MGLinWS mglinws at aol.com
Wed Jul 26 22:49:13 EST 1995


With regard to ensemble (i.e., multiple single-unit) activity, you might
try looking into the statistical literature for methods for pattern
recognition and classification.  Specifically, look for methods that work
with high-dimensional, nonlinear data sets.  In my own research, I have
been recording multi-neuron activty in the basal ganglia and cerebral
cortex of awake, behaving rats during operant tasks. I have used linear
discriminant analysis and a nonparametric method, classification trees
(CART), to examine the realtionship between patterned neural activity and
behavioral 'events'.  I have found that CART works well in identifying
repeating patterns of spike activity, distributed over 10-30 neurons in
the striatum, that repeatedly occur during specific behavioral acts.  CART
seems superior to the LINEAR discriminant analysis, likely due to issues
of robustness, resistance, and the capacity to handle many small bins of
spike counts (e.g., 25 msec).  Other methods that I am currently trying
out include flexible discriminant analysis, a nonlinear form of LDA
developed by Hastie et al., and some neural network methods for pattern
recognition.  

For correlations, try the joint post-stimulus time histogram method and
gravity methods developed by Gerstein and collegues.

For single units and mutual information, check out a book by Krippendorff
on the use of info theory for data analysis.  Also, papers by Richmond's
and Rolls's labs.

For an intro to some of these methods, see my posters at the 94 and 95 SFN
meetings and the STRANGER home page. If you have any questions feel free
to e-mail me.

Mark Laubach
Bowman Gray School of Medicine
laubach at biogfx.bgsm.wfu.edu



More information about the Neur-sci mailing list