Dear Hussein, I would advise you to fit two straight lines using the subset of the data for which they are relevant. After that you could consider fitting a curve to join the straight lines or to approximate the transition between these two states. There are different options for this but I would suggest that this is a secondary issue.

I've looked up some more information. It seem I'm trying to do non-linear regression on my data.

Possible functions I should fit to are:

Linear - there is an obvious deviation from a linear, which is what I'm trying to emphasize.

Polynomial - polynomial equations do not have finite limits which does not fit what I want.

Sigmoid - the data does not appear S-shaped though.

Scatchard - I want the equation of the curve, not a straight line. Could this be transformed into a curve?

Michaelis-Menten - this is probably correct, but normally describes enzyme activity?

Is there possibilities that I have missed?

If any this is correct (particularly the last three), how would I do the regression in Excel?

I use free-ware called essential regression that I find very useful

http://www.oocities.org/siliconvalley/network/1032/CGPage1.html

The following are a couple of excellent articles on how to do non-linear fitting in Excel. These will allow/teach you to use any formula you want for the fit, so you are not limited by build-in formulas.

Kemmer, G. and S. Keller (2010). "Nonlinear least-squares data fitting in Excel spreadsheets." Nature Protocols 5(2): 267-281.

http:\\dx.doi.org\10.1038/nprot.2009.182

Brown, A. M. (2001). "A step-by-step guide to non-linear regression analysis of experimental data using a Microsoft Excel spreadsheet." Computer Methods and Programs in Biomedicine 65(3): 191-200.

http:\\dx.doi.org\10.1016/S0169-2607(00)00124-3

Note that if you find the appropriate formula to fit your calibration curve, you are not limited to the linear part of the calibration curve. Of course the error will be higher the closer you get to the plateau.

Your data will fit well to a Michaelis Menten model, y=a*x/(b+x).This can be easily done in SciDavis or Qtiplot. The parameter a is the saturation point. The parameter a/b is the slope of the initial part of the curve.

I used the Solver add-in for Excel as directed in the two papers given by Johnny Hendriks, and it works very well. For a 'curve' of best fit, I tested a Michaelis-Menten equation, y = a * x / (b + x), and a 4-parameter logistic model, y = (a - d) / (1 + (x / c)b) + d. Both appear to fit well R2 > 0.95. I don't know why, but it seems the Michaelis-Menten equation is trickier to fit, as the resulting equation depends on the initial value for a, the asymptote limit. Maybe limits are hard to find or the data needs more points at signal saturation. Conversely, the logistic, sigmoid curve probably 'overfitted' with 4 parameters compared to only 2 for Michaelis-Menten.

I didn't try the freeware as Solver is bundled with Excel 2003.

To find the linear range, I used linear regression on various subsets of the data and found the largest range with r2 > 0.90.

Thanks

The data should be trivial to fit to the Michaelis Menten equation using good nonlinear fitting techniques, and should be relatively insensitive to initial parameter estimates. Excel is not well suited to this task. I would strongly recommend a good fitting program like Origin, Qtiplot, or SciDavis etc. The last of these is open source.

You do not have sufficient data points to justify a 4 parameter fit.