IntFire4 improvements: thanks to van Elburg and van Ooyen!

The algorithms used in the IntFire4 artificial spiking neuron have been revised to be more biologically plausible, in light of findings described in a paper by Ronald van Elburg and Arjen van Ooyen which appeared in a recent issue of Neural Computation (see REFERENCE below). We would like to take this occasion to describe their results and thank them for making this valued contribution to the community of NEURON users.

IntFire4 is a "current-based" integrate and fire mechanism which has an "excitatory current" that decays with time constant taue, and a biphasic "inhibitory current" described by time constants tau1 and tau2. These currents are integrated by a leaky membrane with time constant taum. The original implementation of IntFire4 required that taum < taue < tau1 and tau2 so that the time of the next spike could be determined by a sequence of event-driven Newton iterations.

van Elburg and van Ooyen proved that these constraints were much too restrictive. They generalized the mathematical basis of IntFire4, showing that the event-driven Newton iteration scheme could reliably find spike times in an integrate-and-fire cell that has any number of monoexponential excitatory currents with different time constants taue0, taue1 . . . and biexponential inhibitory currents with different time constants taui10, taui20, taui11, taui21 . . . , as long as the slowest excitatory current decays faster than the fastest decaying inhibitory current.

They notified us of this important result well before their paper appeared in print, and the IntFire4 source code was revised accordingly in late 2008. Since then it has allowed any combination of taum, taue, taui1 and taui2 (where taui1 < taui2), as long as taue < taui2. The number of time constants remains limited to one exctatory and one pair of inhibitory time constants, however. Users who need multiple different taue's or taui1's and taui2's can use multiple instances of IntFire4, each with a different set of time constants. That said, there is nothing to stop the daring from copying intfire4.mod from nrn-x.x/src/nrnoc/intfire4.mod to a new mod file and making whatever changes they like to its source code--and then debugging it themselves.

REFERENCE

van Elburg, R.A.J, and A. van Ooyen, A. Generalization of the event-based Carnevale-Hines integration scheme for integrate-and-fire models. Neural Computation 21:1913-1930, 2009. Source code available from ModelDB http://senselab.med.yale.edu/modeldb/ via accession number 115357.