This site uses cookies. By continuing to use this site you agree to our use of cookies. To find out more, see our Privacy and Cookies policy. Close this notification
NOTICE: There are currently some performance issues with IOPscience, which may also cause error messages to appear. Apologies for the inconvenience..
The following article is Open access

A Frank mixture copula family for modeling higher-order correlations of neural spike counts


Published under licence by IOP Publishing Ltd
, , Citation Arno Onken and Klaus Obermayer 2009 J. Phys.: Conf. Ser. 197 012019



In order to evaluate the importance of higher-order correlations in neural spike count codes, flexible statistical models of dependent multivariate spike counts are required. Copula families, parametric multivariate distributions that represent dependencies, can be applied to construct such models. We introduce the Frank mixture family as a new copula family that has separate parameters for all pairwise and higher-order correlations. In contrast to the Farlie-Gumbel-Morgenstern copula family that shares this property, the Frank mixture copula can model strong correlations. We apply spike count models based on the Frank mixture copula to data generated by a network of leaky integrate-and-fire neurons and compare the goodness of fit to distributions based on the Farlie-Gumbel-Morgenstern family. Finally, we evaluate the importance of using proper single neuron spike count distributions on the Shannon information. We find notable deviations in the entropy that increase with decreasing firing rates. Moreover, we find that the Frank mixture family increases the log likelihood of the fit significantly compared to the Farlie-Gumbel-Morgenstern family. This shows that the Frank mixture copula is a useful tool to assess the importance of higher-order correlations in spike count codes.

Export citation and abstract BibTeX RIS

Please wait… references are loading.