MultupleRunFitter[0]Generators output differs from sims

[crepatas]

I'm optimizing a NMDA glycine channel. I have 4 curve responses at different Glycine concentrations. At the end of the optimization I get quite a good fit within the multipleRunfitter window. But when I try to run the same simulation outside the runFitter I get quite a different response: in particular inside runFitter I get a current peak around 20ms, while during a normal simulations I get ~2ms (which mean the rising time has not been optimized).

I posted all NEURON windows relative to a Glycine concentration of 100 here:
http://pg.photos.yahoo.com/ph/crepatas/ ... cb&.src=ph
In particulat CI.jpg should have the same form as the black curve in OPT.jpg. (as you can see I set the correct value of c.gly at 100 just like within the MultipleRunFitter)

Any suggestions/comments?

Crepatas

[ted]

when I try to run the same simulation outside the runFitter I get quite a different response

Then the assumption that you have "run the same simulation" is false, i.e. an instance
of mismatch between what's in your head and what's in the computer. Such mismatches
are often cases of the "right under my nose" phenomenon: the error that the author can't
see despite multiple re-readings, but which a pair of fresh eyes can detect fairly quickly.

If you would like me to try to diagnose the problem, please zip up the necessary hoc and
ses files and send them as an email attachment to me at
ted dot carnevale at yale dot edu
If you are using Outlook or Outook Express, please be careful not to send it as winmail.dat,
a proprietary Microsoft format that appears to have been designed only to make email
inconvenient for non-MSWin users.

[ted]

Actually the false assumption was that the MRF's generator was plotting the full time
course of the fitted variable.

The simulated "variable to fit" is captured with Vector.record. After a run, this recording
is interpolated to find the values of the "variable to fit" at the same times as the sampled
data points. The MRF generator's plot shows only the sampled data points, and the
corresponding interpolated "variable to fit" points.

In the example provided by crepatas, there were no sampled data points between 0 and
20 ms. Consequently, no values of the "variable to fit" were plotted between 0 and 20 ms.
Instead of a nice smooth curve, the plot showed a straight line that jumped from the value
at 0 ms to the value at 20 ms. Copying the points that _were_ plotted to NEURON's
clipboard and pasting them into a Vector display tool revealed that these points lay right
on top of the smooth curve that results when the full time course of the "variable to fit" is
plotted. Thus the discrepancy was only an apparent discrepancy.