Fitting steep dose response curves

Question

Hi I'm trying to fit a rather steep dose-response curve in the drc package in R. Moles is my x function and pct_control is my y function. I tried fitting it on the Log-logistic 3 parameter function and the fit was terrible. I tried other model fits and the closest fit I could get was logistic (5 parameter). Anybody have any other ideas of model fit?

 dm2005.test1<-drm(pct_control~moles, data=DONMON,fct=LL.3())

Error in optim(startVec, opfct, hessian = TRUE, method = optMethod, control = list(maxit = maxIt, : non-finite finite-difference value [3] Error in drmOpt(opfct, opdfct1, startVecSc, optMethod, constrained, warnVal, : Convergence failed

moles   pct_control
0.00344408  7.124090958
0.00344408  8.491257546
0.00344408  7.091147185
0.00344408  5.625149276
0.00344408  10.12197432
0.00344408  7.955921231
0.00344408  7.3794052
0.00344408  9.463098856
0.00344408  3.747354203
0.00344408  4.61212825
0.00344408  5.180408338
0.013748484 6.061654272
0.013748484 9.784300645
0.013748484 9.430155083
0.013748484 5.394542864
0.013748484 7.824146139
0.013748484 7.947685288
0.013748484 5.691036823
0.013748484 8.079460381
0.034387728 1.523649511
0.034387728 2.001334223
0.034387728 2.264884409
0.034387728 2.388423558
0.034387728 2.124873372
0.034387728 1.573065171
0.034387728 1.99309828
0.034387728 0.831830274

Good Eats · Answer 1 · 2017-09-29T16:59:46.590

I tried to use your drm code to fit the data provided and did not get the error you obtained. I was able to plot the result and I include that image below. I think the difficulties you are experiencing could be due to the low number of "moles" points -- you have 3. In my experience, if you want to fit a reasonable curve you will need ~5 points that span the lower asymptote, the upper asymptote and the transition region. The reason for this is that with such small coverage, there are many ways to fit the curve. For example, given the data, it is unclear whether the plateau at 7 "pct_control" represents the lower asymptote (e.g. the full dose response would span 100% down to 7%), or if it is the upper asymptote (e.g. the full dose response would span 7% down to 0%) -- or some intermediate result.

Fitting steep dose response curves

1 Answers1