(Sorry about this being a *long* posting, but I've got a lot to ask...:)
I am a 6th year graduate student in the Department of Chemistry and Bio-
chemistry at the University of Texas at Austin. My advisor is Dr. Robert
E. Wyatt. I passed my final defense last semester on November 4th and I
am now working on dissertation revisions.
My research is in the field of Computational Neurobiology and I've spent
the last 5 1/2 years performing 'circuitry studies' of a computational
model of the primary visual cortex (area 17). The modelling programs
(ANATOMY and PHYSIOLOGY) were given to me to begin with and were developed
before I arrived by Dr. Wyatt, a postdoc of his (Dr. Paul Patton) and
another graduate student (Dr. Elizabeth Thomas). My work has been com-
pleted, but my committee has recommended publication of my dissertation as
a monograph and I am currently working on finishing it up for this purpose
and for submission to the University this semester with Dr. H. Grady Ry-
lander III -- a Biomedical Engineering professor who was on my disserta-
tion committee.
Since this work is probably only going to be read by people in the fields
of Neural Networks and Neuroscience in general, I have been asking for
advice on how to deal with a few points in the last chapter that I need to
prove some way from researchers and students in these fields: (1) logical
arguments, (2) a reference somewhere, or (3) revising the model appropri-
ately and proving or disproving my points. Dr. Rylander has called this
an 'error analysis' of the programs. I am only evaluating the modelling
programs which were given to me -- my programs are clear and are basically
useable (or changeable for use) on any modelling program that might come
along. Drs. Wyatt, Thomas, and Patton tabulated some points about the
original programs relating to what they neglected and what they included.
I have taken either these points (or some variation on them) and I have
asked the question of what would happen if one were to either add what was
neglected or delete what was included. I have given these arguments of
mine to these questions to Dr. Rylander and he now wants more references or
proof of the points I have made. I am not sure where I can gather informa-
tion on these points and I was hoping that y'all might be able to give me
either specific references or some direction to go in when it comes to these
points. Also, if they are points that will *have* to be dealt with through
experiments with the programs, please inform me of this. I am trying to cut
down on the number of programs that are involved in proving these points.
I have already put together methods for proving the points I have listed if
this *must* be done, but I would prefer to avoid this somehow with refer-
ences of some kind.
Here is a tabulation of the points that I am looking at and the associated
information that I already have. Some of the information I have added on
the EFFECT OF INCLUSION (NEGLECTING) IN THE MODEL is still somewhat unclear
and needs to be revised a bit. It is what I've come up with upon thinking
about the model and what data I have gathered as to how these factors would
change things:
-----------------------------------------------------------------------------
DETAIL IN ARCHITECTURE
-----------------------------------------------------------------------------
POINT #: 1
FACTOR NEGLECTED IN MODEL:
Only dLGN afferents (no corticocortical input fibers).
EFFECT OF INCLUSION IN MODEL:
This would increase the number of connections in the model due to all of
the extra input coming in, which would complicate the circuits obtained.
(Heard this is related to some type of 'decorrelation'. Any references?)
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Add more afferents and this would become obvious.
-----------------------------------------------------------------------------
POINT #: 3
FACTOR NEGLECTED IN MODEL:
Cells on lateral edges neither transmit nor recieve signals beyond the
cortical box.
EFFECT OF INCLUSION IN MODEL:
In order to include this and for it to make sense, the cortical slab
would have to be enlarged and other areas of input would have to be
included. Another point that really doesn't mean too much in the case
of my research...
HOW TO GET THE DATA NEEDED IN PROOF:
Can't really prove this one -- just makes sense logically -- so NOT
NEEDED!!!
-----------------------------------------------------------------------------
POINT #: 6
FACTOR NEGLECTED IN MODEL:
Randomness, rather than developmental rules, are used to assign soma
locations.
EFFECT OF INCLUSION IN MODEL:
Developmental occurs over a long period of time and the rules are not
known well. If they were, the locations of the cells are still 'random'
from animal to animal. (Cells are 'dropped' into the model at random
spots in distributions as found in the data or as estimated by the
original modellers.)
HOW TO GET THE DATA NEEDED IN PROOF:
Again, can't be proven -- just logical. LIBRARY -- Can probably find
something in the literature about locations of cells being 'random' from
animal to animal...?
-----------------------------------------------------------------------------
DETAIL IN SYNAPSES AND CONNECTIONS
-----------------------------------------------------------------------------
POINT #: 1
FACTOR NEGLECTED IN MODEL:
Fixed, nonplastic connection strengths.
EFFECT OF INCLUSION IN MODEL:
Only a factor if the model is run for a longer period of time.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Could set up program to vary connection strengths with time
according to either some version of Hebb's Rule or some other simplified
method. MY DATA -- Evidence somewhat seen in various data sets with
different connection strength values.
-----------------------------------------------------------------------------
POINT #: 2
FACTOR NEGLECTED IN MODEL:
(Is the difference between somatic and dendritic connections enough?
Should there be little changes due to travelling relatively so fast
and the signal doesn't change relative to the soma of other classes
based on the locations of neurons?)
EFFECT OF INCLUSION IN MODEL:
For most models this is enough. This assures that the connections
matter very much.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- (b) Add a way to vary seed and delay in the model in some
simplified form.
-----------------------------------------------------------------------------
DETAIL IN CELL PHYSIOLOGY
-----------------------------------------------------------------------------
POINT #: 1
FACTOR NEGLECTED IN MODEL:
There are nonspecific membrane channels.
EFFECT OF INCLUSION IN MODEL:
This detail effects the use of the Hodgkin-Huxley equations of the next
section. If this is included, then it makes sense to consider the use
of the Hodgkin-Huxley series of equations.
HOW TO GET THE DATA NEEDED IN PROOF:
LIBRARY OR PROGRAM -- Find more information on H-H equations and either run
them or find data in articles in support of this.
-----------------------------------------------------------------------------
POINT #: 2
FACTOR NEGLECTED IN MODEL:
The voltages are not found in axonal or dendrtic components, just a single
voltage for the whole cell.
EFFECT OF INCLUSION IN MODEL:
This detail wouldn't contribute that much if the original voltages that
were used for the overall cell were reasonable.
HOW TO GET THE DATA NEEDED IN PROOF:
LIBRARY -- See more references using this kind of model method and analyze
their data. PROGRAM -- Also use those programs (articles) to vary model
and test this out.
-----------------------------------------------------------------------------
POINT #: 4
FACTOR NEGLECTED IN MODEL:
There are uncertainties about conduction velocities in axons.
EFFECT OF INCLUSION IN MODEL:
This detail would change things extensively if the uncertainties are large
about the conduction velocities, but hopefully they are not.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Could also vary conduction velocity ranges and see what happens
to the model overall.
-----------------------------------------------------------------------------
POINT #: 5
FACTOR NEGLECTED IN MODEL:
A fixed firing threshold and bursting patterns, which are independent of
firing history, have been used.
EFFECT OF INCLUSION IN MODEL:
This detail would be important if one were to run the model for a longer
period of time than is being used here. Those values would change with
time, but how *much* time?
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Vary these factors over time (longer running model?) and see
what happens to the data.
-----------------------------------------------------------------------------
LESS DETAIL IN CELL MORPHOLOGY
-----------------------------------------------------------------------------
POINT #: 1
FACTOR INCLUDED IN MODEL:
Stick neurons with three-dimensional axons and dendrites.
EFFECT OF NEGLECTING IN MODEL:
Important in getting correct connections.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Could lessen the details of cell morphology and see what happens.
-- Could also use simple cells of somas and very little or nothing else,
but in large numbers and see what happens.
-----------------------------------------------------------------------------
LESS DETAIL IN ARCHITECTURE
-----------------------------------------------------------------------------
POINT #: 1
FACTOR INCLUDED IN MODEL:
17 types of cells (10 excitatory and 7 inhibitory) as described in
Table A.1.
EFFECT OF NEGLECTING IN MODEL:
Would change the patern of connections found in the model. Specific
connection types would no longer be known and the circuits would be
more generalized.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- No longer use specific models of each layer cell type, but one
generalized one for each cell type and see what happens overall.
LIBRARY -- Also, could examine Thomas, Patton, and Wyatt's data with
respect to this when they tried generalized cell types of limited struc-
ture variation and non-specific naming.
-----------------------------------------------------------------------------
POINT #: 2
FACTOR INCLUDED IN MODEL:
The distribution of cell types in the cortical layers is according
to experimental values.
EFFECT OF NEGLECTING IN MODEL:
Would change the pattern of connections found in the model. Specific
connection types would no longer be known and the circuits would be more
generalized.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Change distributions around and look at the effect.
-----------------------------------------------------------------------------
POINT #: 4
FACTOR INCLUDED IN MODEL:
There are both X and Y afferent fibers.
EFFECT OF NEGLECTING IN MODEL:
Distribution of these is carefully considered in the model. If this was
neglected, then stimulation would occur, but not a properly distributed
sites. This would change the circuits obtained.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Use generalized fibers without extra details of each afferent
type and see what happens.
-----------------------------------------------------------------------------
LESS DETAIL IN SYNAPSES AND CONNECTIONS
-----------------------------------------------------------------------------
POINT #: 1
FACTOR INCLUDED IN MODEL:
Excitatory and inhibitory connections.
EFFECT OF NEGLECTING IN MODEL:
Important to keep the connections right and for the model to work
properly.
HOW TO GET THE DATA NEEDED IN PROOF:
LIBRARY -- Logical. Also, if all excitatory then like epileptic seizures
(reference?). If all inhibitory, won't work -- no activity (reference?).
-----------------------------------------------------------------------------
POINT #: 2
FACTOR INCLUDED IN MODEL:
Multiple connections are possible between pairs of neurons.
EFFECT OF NEGLECTING IN MODEL:
This allows for detail in synaptic connections amongst cells to get the
properties right related to activation or not. Essential in determining
whether a particular cell will be activated by the stimulation it receives,
i.e., whether enough cells have stimulated it enough. If this factor were
left out, the connection strength would become especially important.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Use single connections in program and this will be proven.
Rather logical, so not sure *why* this is needed...?
-----------------------------------------------------------------------------
POINT #: 3
FACTOR INCLUDED IN MODEL:
Dendritic and somatic connections are possible.
EFFECT OF NEGLECTING IN MODEL:
If eliminate one or the other, the questions would be simplified. This
effect is interrelated to the factor of including separate dendritic
and somatic synaptic delay times. (See LESS DETAIL IN DYNAMICS - #4)
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Drop one or the other and change equations acordingly and
see what happens.
-----------------------------------------------------------------------------
POINT #: 4
FACTOR INCLUDED IN MODEL:
Cell types have specifc targets.
EFFECT OF NEGLECTING IN MODEL:
Could be dropped and go by where connections occur, but leads to problems
with exceptions: basket and chandelier cells. All somatic connections to
these cells are converted to dendritic connections as the model is
programmed.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Eliminate exception rules and see what happens.
-----------------------------------------------------------------------------
POINT #: 5
FACTOR INCLUDED IN MODEL:
There is a large fan-in and fan-out due to three-dimentional axonal and
dendritic arborization.
EFFECT OF NEGLECTING IN MODEL:
Results of 2, 3, and 4 above. If changes are made, then this would change.
HOW TO GET THE DATA NEEDED IN PROOF:
Goes along with above so NOT NEEDED!! (Can this point be ommitted?)
-----------------------------------------------------------------------------
LESS DETAIL IN CELL PHYSIOLOGY
-----------------------------------------------------------------------------
POINT #: 1
FACTOR INCLUDED IN MODEL:
Neurons operate in either a graded or an impulse mode for realism.
EFFECT OF NEGLECTING IN MODEL:
Extends out the activation time of a cell which leads to a more proable
chance of the next cell being activated. If eliminated, this could reduce
the number of cell connections in the circuits.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Eliminate a mode and see what happens.
-----------------------------------------------------------------------------
POINT #: 2
FACTOR INCLUDED IN MODEL:
Axonal propagation time is included.
EFFECT OF NEGLECTING IN MODEL:
Important because axons carry signals at a slower rate than dendrities.
Adds reality to the timing of the patterns of the connections.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Drop and see what happens.
-----------------------------------------------------------------------------
POINT #: 3
FACTOR INCLUDED IN MODEL:
Firing threshold (single or multiple-spike) and post-firing hyperpolariza-
tion have been included.
EFFECT OF NEGLECTING IN MODEL:
Important because one can not have a large group firing at the same time,
as found in epilepsy. Must be more cells or all controlled in order to
process data properly.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Drop these and see what happens.
-----------------------------------------------------------------------------
LESS DETAIL IN DYNAMICS
-----------------------------------------------------------------------------
POINT #: 1
FACTOR INCLUDED IN MODEL:
When cell activity is below threshold integration for the system of non-
linear, differential-difference equations for cell voltages, Vi(t), is
done.
EFFECT OF NEGLECTING IN MODEL:
Important for making the model as realistic as possible and to connect the
activities of all of the cells to one another so they are working as a
synchronous unit.
HOW TO GET THE DATA NEEDED IN PROOF:
LIBRARY -- Rather logical and makes sense based on the physiology presented
in other models and also the H-H equations. (References in proof of this?)
-----------------------------------------------------------------------------
POINT #: 4
FACTOR INCLUDED IN MODEL:
Axonal and synaptic delay times are included.
EFFECT OF NEGLECTING IN MODEL:
The effect would be limited if possibly an average synaptic delay were
used instead of separate ones for dendritic and somatic.
HOW TO GET THE DATA NEEDED IN PROOF:
PROGRAM -- Try eliminating these and using an average one instead of two
different ones and see what happens.
-----------------------------------------------------------------------------
Please go over these and tell me what you think and if you happen to know
of any references.
Thanks,
Eva S. Simmons
--
************************ THE SIMMONS FACTOR **********************************
Eva Sabrina Simmons University of Texas at Austin Ph.D. (new)
Theor P-Chem -- Field: Computational Neurobiology weevey at dopey.cc.utexas.edu
****************** WATCH IT, OR IT MIGHT ATTACK!! ****************************
