X-Message-Number: 7901
Date: Thu, 20 Mar 1997 08:38:10 -0800 (PST)
From: Joseph Strout <>
Subject: neuron emulation, state of

You can't get anything past Thomas, can you?  :)  It's true, the current
state of the art in neuron emulation is crude at best.  Here's where we
are, at least as I can see it:

1. We can emulate a neuron's spike-generating function.  That is, given
the current at the spike initiation zone (usually the cell body), we can
predict when the spikes will occur.  NOT just the statistical properties
of the spike train, but exactly when each and every spike will happen.
This is usually done via kernel-based "black box" models, which can be
configured in real time.  That is, some folks have connected a neuron (via
an electrode, of course) to a computer, and within minutes, the computer
can predict the neuron's spikes as they happen.

2. We can reproduce most of the signal integration that occurs in the
dendritic tree via compartmental modeling.  This takes into account active
ion channels, second messengers, and other good stuff which you might
think is difficult to reproduce.  It's not that difficult, in principle;
it's just a big set of differential equations.  The trouble, of course, is
that the set can't be solved symbolically, and can only be approximated on
a digital computer.  So we approximate, and to get convergence (i.e., to
keep the results close to the "correct" answer over extended periods of
time), we use lots of compartments and very small time steps.  It's a
hideously slow process, but it does a fair job.

3. A few groups have started using analog VLSI to model neurons on
microchips.  To simplify their engineering, they strip away most or all of
the dendritic properties listed as #2 above, and basically build a network
of black boxes of the sort described in #1.  But it's a start.  These
chips are very fast.

Hmm, these are rather arbitrary distinctions, as there are a number of
models that combine various levels.  For example, many researchers look at
small networks of neurons with simplified (but present) dendritic trees.
Nobody has modelled more than a neuron or two in excrutiating detail,
because the computation involved is overwhelming.  But we're getting
there!

Cheers,
-- Joe

,------------------------------------------------------------------.
|    Joseph J. Strout           Department of Neuroscience, UCSD   |
|               http://www-acs.ucsd.edu/~jstrout/  |
`------------------------------------------------------------------'

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=7901