Hi Patrick,
this is not a solution to your problems with getting 'nls' to work, but
may still be of some use to you.
The package 'drc' on CRAN offers an alternative to 'nls'. See
the
following example using your data frame mydata2.
library(drc)
## Defining the non-linear function
"fct1" <- function(dose, parm)
{
100 * exp(parm[,1]+parm[,2]-parm[,3]) *
(exp(-exp(parm[,1])*dose)-exp(-exp(parm[,2])*dose))
/(exp(parm[,2])-exp(parm[,1]))
}
## Fitting the model (for details see ?multdrc)
model1 <- multdrc(Conc+1~Tps, Organ, data=mydata2, fct=list(fct1, NULL,
c("lKe","lKa","lCl")), startVal=c(-2.77, -2.77,
-2.77, -1.41, -1.41,
-1.41, -1.13, -1.13, -1.13))
summary(model1) # a lot of parameters could be 0!
plot(model1, log="")
## Plots for each level of Organ
plot(model1, level="Carc", ylim=c(0,5), log="")
plot(model1, level="Foie", log="")
plot(model1, level="TD", ylim=c(0,6), log="")
## Fitting the model with a transform-both-sides Box-Cox transformation
## - in an attempt to adjust for variance inhomogeneity (visible in the
residual plot)
model2 <- multdrc(Conc+1~Tps, Organ, data=mydata2, fct=list(fct1, NULL,
c("lKe","lKa","lCl")), startVal=c(-2.77, -2.77,
-2.77, -1.41, -1.41,
-1.41, -1.13, -1.13, -1.13),boxcox=T)
summary(model2)
Hope this is of help to you!?
Best regards
Christian