Gaussian noise

Moderator: wwlytton

Post Reply
Sandrine

Gaussian noise

Post by Sandrine » Wed May 14, 2008 9:09 am

I have an entire detailed (with compartments) cortical column model with 3 layers. For the moment, the column is isolated. So in order to be more realistic, i would like to represent the "rest of the world" of my column by Gaussian noise (mean =0) that i would apply in each of my neuron.
Is it an easy way to do that in hoc?

Thank you for your help,

Sandrine.

ted
Site Admin
Posts: 5570
Joined: Wed May 18, 2005 4:50 pm
Location: Yale University School of Medicine
Contact:

Re: Gaussian noise

Post by ted » Thu May 15, 2008 10:39 am

Sandrine wrote:Is it an easy way to do that in hoc?
To do this efficiently, you should use a mechanism specified in NMODL--a point process if
you only want to deliver current to one location, or a density mechanism if you want to
deliver current throughout the entire cell.

Before creating something brand new (which also means testing and debugging it), you
might want to try something that already works. Alain Destexhe developed a noise source
that emulates fluctuating synaptic conductances--"more realistic" than just fluctuating
current--and used it in several publications
Destexhe A, Rudolph M, Fellous JM, Sejnowski TJ (2001)
Fluctuating synaptic conductances recreate in vivo-like activity in neocortical neurons.
Neuroscience 107:13-24
Rudolph M, Destexhe A (2005)
An extended analytic expression for the membrane potential distribution of
conductance-based synaptic noise.
Neural Comput 17:2301-15
Their mod file is available from ModelDB as Gfluct.mod in
http://senselab.med.yale.edu/ModelDB/Sh ... odel=64259
It is nicely parameterized, so you can specify the mean inhibitory and excitatory conductances
and their associated reversal potentials.

There is one problem with Gfluct.mod: the conductances are sampled from a Gaussian
distribution and therefore may become negative. This is a nonsense result that could
distort the outcome of simulations. It is possible to trap negative values (use an "if"
statement to replace negative values with 0). This was done by Migliore et al. (2006) and
Hines et al. (2008)
Migliore M, Cannia C, Lytton WW, Markram H, Hines ML (2006)
Parallel Network Simulations with NEURON. J Comp Neurosci 21:110-119
Hines M, Eichner H, Schuermann F (2008)
Neuron splitting in compute-bound parallel network simulations enables runtime scaling
with twice as many processors J Comput Neurosci in press.
who modified Destexhe's code and called it Gfluct2.mod (go to
http://senselab.med.yale.edu/ModelDB/Sh ... odel=97917
and look inside the pardentategyrus directory).
Trapping negative conductances distorts the statistics of the simulated fluctuating
conductance, but it is more physioilogical than allowing negative conductances
associated with either excitatory or inhibitory reversal potentials (which would produce
synaptic currents with the opposite effect).

Regardless of whether the noise source is a current source or uses fluctuating
conductances, you will want to verify that it has no effect on mean membrane potential.
Even if mean noise current is 0, when applied to a cell with active membrane properties
it may still induce a shift of mean membrane potential. To ensure that this does not
happen, you should set all synaptic weights to 0 and run a test simulation. The test
simulation should last sufficiently long that membrane potential, gating states, and ionic
concentrations can settle to stable mean values. For each cell, calculate mean
membrane potential at the site of noise injection over a "sufficiently long interval" near
the end of the test simulation. If mean membrane potential has shifted from rest (or
whatever you think should be the cell's "mean operating potential"), estimate the
adjustments to the noise that you think will correct this shift, apply the adjustments,
and run another test simulation. Repeat until you are satisfied with the results.

Post Reply