>
> General comment:
> Why has Harry gone through the trouble of creating another modeling
> software program?
Thanks for the comments. The reason for building the tool from scratch was
not with the intention of continuing to develop it in competition with the
existing families of tools, but rather to get detailed insight into the
real issues using a relatively hardware-independent program before I
downselected to one of those existing tools (and get locked into the
corresponding hardware architecture). After I finished the basic program
on a SGI target/Mac IIfx development environment, I rehosted to a PVM3
environment and ran into a new collection of issues--that experience was
valuable, too. I'd recommend that any graduate student planning to work in
neural modeling build his or her own model first as a throw-away before
using existing tools.
>
> Comment 5
"Synaptic release models in most published compartmental models are (to
be polite about it) simplistic, typically strongly influenced by the
artificial neural network model of the spiking neuron."
> This is a legitimate issue. Some solutions to this problem have been
> suggested in the literature (e.g. E. De Schutter, J.D. Angstadt and R.L.
> Calabrese: A model of graded synaptic transmission for use in dynamic
> network simulations. Journal of Neurophysiology 69: 1225-1235, 1993),
> but general implementations are lacking. Electrical synapses have
> been implemented in GENESIS in a manner compatible with the implicit
> solution methods.
Thanks for confirming a suspicion. Trying to model the GABAergic synapses
in the olfactory bulb when there was so little in the literature to go on
was really the most frustrating aspect of the study. I had done earlier
work that used Katchalsky network models, and translating that into more
biologically-realistic terms turned out to require models that didn't
exist. Even the available models for excitatory synapses were cruder than
I needed. And, yes, I suspect that GENESIS is the tool of choice for what
I'm planning to do over the next two years.
>
> Comment 6
"The crucial role of active tuning..."
> I would advise Harry to be very careful with such a statement. This
> might very well reflect lack of robustness of his model, due to
> simplifications, errors, lacking data... Also did he control for
> accuracy of integration problems?
It's just an interim report... Actually that comment reflects some other
models and analyses, too. My real goal was to get some understanding of
whether detailed biologically realistic models would translate to
Professor Freeman's Katchalsky network models at a high enough level of
aggregation. That is, I was investigating the conditions under which the
Katchalsky network models would be correct. No conclusions, but I'm
suspicious that active tuning has a big role. Note that I found the
synaptic models that I investigated appeared to be too crude to work well
in a model that involves active tuning.
>
> Comment 7-9
> I am not familiar enough with the olfactory bulb to comment on these
> results. But I wonder if the result in comment 7 is an emergent
> property of the simulation, or was the behavior of the periglomerular
> build into the system?
>
I found I had to model shunting inhibition in some detail to get anything
even halfway realistic in the dynamics. It also clarified a disagreement
between Shepherd's and Freeman's descriptions of how the olfactory bulb
functions. Yes, periglomerular cells are GABAergic, but they function
through shunting inhibition. In a variable chloride gradient generated by
a varying activity level of the tufted/mitral apical dendrites, they are
the basis of the Weber-Fechner power law conversion of the sensory signal
(per W. Freeman, 1994). That will not be easy to model, since it involves
keeping track of the extracellular chloride ion concentration.
I still don't really understand the role of the 'H'-type synapses between
the olfactory sensory cells and pairs of tufted/mitral cells and
periglomerular cells, connected by reciprocal synapses. There's an article
in the latest Neural Computation that seems to indicate they may be
involved in 'directionally'-dependent sensitivity, but I'm not sure what
that means in the context of a nose. To get insight, I will have to push
my model quite a distance further. Also the way the granule cells
function is still unclear. In the topology identified by the anatomists,
unless the system is quite sensitively tuned, the granule cells can
provide overwhelming shunting inhibition, preventing most tufted/mitral
cells from even participating in the output of the OB--at marked variance
with experiment, which shows a 'spiral' signal being generated, with a
sequence of t/m populations being involved, and with all eventually
participating. Then again, they might not even activate for low-level
signals, resulting in garbage output from the OB. Yet olfactory granule
cells appear to have only excitatory synapses. That doesn't give much
controllability.
> Erik De Schutter
> Born Bunge Foundation, University of Antwerp, Belgium
> erik@kuifje.bbf.uia.ac.be
>
With thanks for the comments,
Harry Erwin
Internet: herwin@gmu.edu
=============================================================================