I am stimulating my cells in culture using a variety of stimuli and checking the response as Calcium Intensity changes, using live cell calcium imaging.
The rise is instantaneous but decay seems exponential.
I am fitting this decay using exponential decay model in MATLAB and getting an R-square estimate.
What R - square value is considered a good fit ?
Should i discard the exponential decay curves, below a certain value of R-Square ?
You may use non-linear fitting for least-squares minimization procedure. Set R-square to be “1.0”. If the fitting is not satisfactory, they you may need to change to two-exponential function.
Hello sir !
Thanks for the reply
Acutally !! That's what i am doing..
Non linear curve fitting using least squares, yet some of the data sets have a low R-square value (say R-square less than 0.9). Even changing it to two exponential does not change the R-square much...
I am doubtful about the value of R-square, which i can consider as a good fit, ?
The R squared value is a measure of how well the dependant variable (decay curve) can be predicted from your independent variables. At 1.0 there is a 100% match at 0.9 , 90% , 0.8 80% etc.
Before using a cut-off it Would be prudent to ask why the R2 values are low. Could these decays be biphasic and be better suited to a double exponential fit or is there a lot of noise in the traces that could be filtered? Maybe only using part of the curve could yield a more accurate fit.
failing the above a cut-off applicable to your data would be sensible, where you neither lose too much data nor have too much noise stopping you determining true differences in experimental groups. Unfortunately this is not a set value.
good luck
For your reference (see attached) regarding non-linear fitting procedure
thanks George W P Madders for your valuable comment
i have tried with bi-exponential model too and that has yielded better results (higher R-square) in few data sets
and it seems to me that the low R-square values are due to noise in the data, resulting in poor fitting
i don't know whether smoothing will improve the fit or not, haven't tried yet
And i second you on the point the R-square is not a set value ... :)
Dear Tushar,
I share Madders opinion. It would certainly be convenient to determine whether or not your data can be fitted by more than one exponential instead of forcing fitness to a single one.
- Badges
- Science topic