From: Dieter Menne <dieter.menne_at_menne-biomed.de>

Date: Tue, 08 Apr 2008 17:05:12 +0000 (UTC)

Time = seq(0,2,by=0.1)

plot(b$Time,b$RFU)

lines(Time, A + (B+c*Time)/(1+exp(k*(e-Time))))

R-help_at_r-project.org mailing list

https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. Received on Tue 08 Apr 2008 - 17:07:47 GMT

Date: Tue, 08 Apr 2008 17:05:12 +0000 (UTC)

jarod_v6 <jarod_v6 <at> libero.it> writes:

*>
*

> I have that problem:

*> this is file : PBS.txt
**> Time RFU
**> 0.0000 27.3021
**> 0.0800 26.1565
*

....

> rm(list=ls())

*> print("ls()")
**> #carica Dati con file di testo
**>
**> b <-read.table("PBS.txt", sep="\t", header= TRUE)
**> print("b")
**> b
**> nlmod1 <- nls(RFU ~ A + (B+c*Time)/(1+exp(k*(e-Time))),
**> data = b,
**> start = list(A = 1, c = 1, B = 1, k =1, e = 1),
**> trace = TRUE))
*

Thanks for providing a self-running example. There are two reasons for the lack of success. First, your starting values were totally off (the curve was rising strongly). Always plot the curve with the starting values together with the data to be fitted, to be sure, that your are at least coming close. Note that your data level off at around 22, so having A=22 is a good first choice.

With the starting values shown below, the iteration starts quite well, but ends in a "step factor reduced ..." error. This almost always means: you want to much from your data. The algorithm cannot decide, what combination of the FIVE coefficients you ask to be estimated is the right one. You have a very smooth curve with little "personality", so I would never use more than 2 or 3 coefficients to fit these. Or, less polite: your model is nonsense; try a simpler one, for example as shown below. And in case you are sure that your model is "the only truth": if you have a whole set of curves, for example from a series of pharmacological tests, you might try nlme; it may work, but not from the beginning.

In case you should argue that software XX does give results (I know a few packages that do): these are wrong. If you are lucky, the software gives you the standard error of the coefficients, possibly in the order of 10^7, which should make you think it over.

Dieter

b <-read.table("PBS.txt", sep="\t", header= TRUE)
print("b")

b

A = 22 c = -0.1 B = 10 k = -2 e = 0

Time = seq(0,2,by=0.1)

plot(b$Time,b$RFU)

lines(Time, A + (B+c*Time)/(1+exp(k*(e-Time))))

nlmod1 <- nls(RFU ~ A + (B+c*Time)/(1+exp(k*(e-Time))), data = b,start = list(A = A, c = c, B = B, k =k, e = e), trace = TRUE)

nlmod2 <- nls(RFU ~A - B*exp(k*Time),

data = b,start = list(A = A, B = B, k =k),
trace = TRUE)

lines(b$Time,predict(nlmod2),col="red")

R-help_at_r-project.org mailing list

https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. Received on Tue 08 Apr 2008 - 17:07:47 GMT

Archive maintained by Robert King, hosted by
the discipline of
statistics at the
University of Newcastle,
Australia.

Archive generated by hypermail 2.2.0, at Tue 08 Apr 2008 - 17:30:28 GMT.

*
Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-help.
Please read the posting
guide before posting to the list.
*