NEURON vs GENESIS - Opinions, please?

mat mats_trash at hotmail.com
Sun Jun 30 12:45:47 EST 2002


> The problem came when I wanted to -add- a neuron. I essentially had to
> rewrite the code from scratch. This is probably the -main- advantage
> of using a canned program. They've already implemented a built in
> "scalability" (especially Genesis).
> 
> On the other hand, figuring out your own solution to the scalability
> problem would sure teach you a lot too.

:) As always with these things I began thinking that it would be
reasonably straightforward affair to encode the number and branching
structure of neurons along with relevant global and compartmental
parameters in a few matrices which I would then run through a
integration algorithm.  I had not really considered computational
efficieny at the outset  but this seems to be a 'big thing' to put it
mildly.  Looking through some of the documentation for the
aforementioned simulators it seems that encoding the dendtritic
arborisation in matrices and then reducing this to a tridiagonal
matrix is at the very heart of both neuron and genesis, but I'm
unclear as to how exactly the structure is "encoded" in the matrix. 
I've been looking for the original paper by Hines (1984) but it seems
to have been published in a rather obscure journal.

Can anyone point me to other sources explaining the Hines matrices
etc.  The way I was thinking of doing it was a much more explicit
almost compartment-object approach, but this intuitively seems to
involve a much greater cpu overhead.  Also, does anyone know of good
sites dealing with the finer points of numerical integration
techniques, especially implicit.

> 
> Another option with Matlab is that you can compile your own code, then
> call that with Matlab. I find Matlab really great for writing GUIs, so
> you could have a Matlab GUI front end for your simulator, the
> simulator as compiled code, and a GUI backend for analyzing the
> output. That would be efficient, because solving the diffeqs will be
> the really cpu-intensive part so you would want that compiled. But you
> may not want to write a whole GUI in fortran (can fortran even do a
> gui?).

This was essentially my idea too, to write the minimal amount of code
required for the core integration routines and then analyse the data
in something like Matlab.  You can buy third-party routines to create
fortran GUIs (e.g. GINOMENU), but I personally do not really want to
spend much time doing that as I reckon a simple command-line interface
is just as powerful and probably easier to implement, though I have no
experience of creating GUIs in Matlab.
> 
> 
> Oooh. This sounds like fun...

Should be, hopefully.



More information about the Neur-sci mailing list