2

Can pgfplots do nonlinear curve fitting?

In Excel, the fitting option is called log(agonist) vs response.

I couldn't find an option in the regression section of pgfplots.

jub0bs
  • 58,916
dustin
  • 18,617
  • 23
  • 99
  • 204
  • PGFPlots can do logarithmic and semilogarithmic lines of best fit, but not general nonlinear regression (i.e. if you need other transformations, you'll have to transform the data yourself. This can be done on the fly in PGFPlots) – Jake May 14 '14 at 04:06
  • @Jake on wikipedia about Dose response, it says: 'Statistical analysis of dose-response curves may be performed by regression methods such as the probit model or logit model, or other methods such as the Spearman-Karber method. Empirical models based on nonlinear regression are usually preferred over the use of some transformation of the data that linearizes the dose-response relationship.' I think transforming the data wouldn't be advised for this. – dustin May 14 '14 at 04:42
  • Wikipedia about nonlinear curve fitting: "Some nonlinear regression problems can be moved to a linear domain by a suitable transformation of the model formulation. ...the choice to perform a nonlinear transformation must be informed by modeling considerations." – Jake May 14 '14 at 04:50
  • @Jake this model is based on Dose response though. – dustin May 14 '14 at 04:51
  • "Log(agonist)" sounds a lot like a logarithmic transform, not logit or probit. – Jake May 14 '14 at 04:55
  • Could you link to a resource that specifies what the excel function does? Or include some sample data and associated excel output? – Jake May 14 '14 at 04:57
  • @Jake here is an explanation on it. You may be better suited at determining if it can be done with the transformation: http://www.graphpad.com/guides/prism/6/curve-fitting/index.htm?reg_dr_stim.htm – dustin May 14 '14 at 04:57
  • @Jake I can get some sample data tomorrow. The data I have right now is nonpublished research data so I cant post that. – dustin May 14 '14 at 04:58
  • @dustin you take the log of the agonist and plot it against the response. So a simple log-transform is what you need. – Trefex May 14 '14 at 08:02
  • @dustin: The log(agonist) vs response curve in GraphPad is a logarithmically transformed sigmoid model (f(x) = a + (b-a) / [1+10^(d*(log(c)-log(x)))]). This model has to be fitted using nonlinear regression, which PGFPlots currently cannot do. You'll have to find the parameters using another program. You could do this using Gnuplot from within PGFPlots. – Jake May 14 '14 at 08:18
  • 1
    Very related: Non-linear curve fitting with gnuplot – Jake May 14 '14 at 08:22
  • @Jake I can delete the post or vote to close based on the link. I think it may be better to close in case some use the same search criteria. – dustin May 14 '14 at 14:59
  • @dustin: I agree – Jake May 14 '14 at 15:02

0 Answers0