In article <3s9d2r$qu0 at vixen.cso.uiuc.edu> Kevin Spencer,
kspencer at s.psych.uiuc.edu writes:
> I'm not aware of any cases (yet) in
>which a model of a neural system predicted certain phenomena that were
>later found empirically.
In defense of modeling, there *are* a number of cases in which models
have predicted phenomena that were later confirmed by experimental tests.
Here are a few examples:
Hodgkin and Huxley described action potential generation in squid axon by
modeling ionic conductances as resistors that changed their resistances
in response to charge movements in the membrane, secondary to changes in
the transmembrane voltage. The theory predicted that currents due to
these charge movements should be measurable. About thirty years later,
these "gating currents" were in fact detected by Armstrong, Bezanilla and
others, and their study has led to a lot of information about ion channel
Patrick O'Hara et al. (Neuron, 11:41) used the sequence and known
structure of bacterial periplasmic amino acid binding proteins to model
the location and structure of the glutamate binding domains of some
mammalian glutamate receptors. These predictions for the metabotropic
glutamate receptor were supported by site directed mutagenesis
experiments. Furthermore, the predictions about the topology of some
glutamate-gated ion channels ran directly against the accepted putative
transmembrane structure. A number of labs, using different techniques,
have recenlty obtained evidence in favor of O'Hara's predictions, forcing
a revision of the accepted notion of GluR topology..
When the "voltage dependence" of the NMDA receptor was first described,
and it's mechanism elucidated, it was proposed that this property suited
the receptor to serve as a sort of coincindence detector (e.g. the
receptor would only operate if postsynaptic depolarization was coincident
with presynaptic transmitter release). This was somewhat reminiscent of
psychological theories about mechanisms of associative memory. It turned
out that this proporty of the NMDA receptor is, in fact, necessary for
many forms of long term potentiation (LTP), the strongest current
neurophysiological theory of memory formation.
Eve Marder's lab has been using the "Dynamic Clamp" (TINS 16:389 and J.
Neurophysiol. 69:992) to model the introduction of novel ionic
conductances into neurons participating in network behavior. This is a
really neat technique. First, they study the existing conductances in a
cell or a number of cells in a network. Then they simulate the network on
a computer. Then they introduce new simulated conductances into modeled
cells to predict what that conductance should do to network behavior.
Finally, they use the dynamic clamp to interactively inject current into
the real neuron in the real network, to simulate the conductance that
they modeled by computer. They have successfully predicted the effects of
a number of conductances on network behavior in this way, and confrimed
these predictions with wet experiments.
I'm sure the list goes on.