Departments of Computer Science and Neurology
NEURON MODEL OPTIMIZATION THROUGH
MINIMIZATION OF RESIDUAL VOLTAGE CLAMP CURRENT
A critical problem at the beginning of neuron modeling is determining whether the conceptual model can reproduce the experimental phenomena; i.e. the discovery of model parameter values which give voltage trajectories in response to current stimuli similar to trajectories recorded in experiments. The success of parameter search methods is dependent on the choice of the function which measures the error or fitness of the model output with respect to the data. Error functions based on the square norm of the difference between model and data trajectories often result in poor optimizer performance because the shape of the function far from the minimum gives no information about the location of the minimum.
Voltage clamp optimization insures that the model never leaves the data trajectory by supplying a residual current that forces the model voltage to be the experimentally recorded voltage.
The model + voltage clamp system has two desirable properties: 1) The nonlinear effect of complex channel gating properties on the membrane potential is removed. Consequently there is no voltage mediated nonlinear coupling between channels. 2) The residual clamp current is linear with respect to the maximum conductances and reversal potentials of each voltage-gated channel. Thus these parameters can be removed from the search space by evaluating them analytically when the error function is calculated.
Unfortunately, minimization of the square norm of the residual current does not yield parameter sets that give the same action potential initiation times as the experimental trajectories because the subthreshold currents that initiate action potentials are much smaller than the currents producing the spike itself. How would one weight the residual clamp current to take into account the importance of these small currents?
Seminar to be held in Room 107, 24 Hillhouse Avenue at 4:15 pm