In lshen1/drexplorer2: Dose-response Explorer2

knitr::opts_chunk$set(
  message = FALSE,
  digits = 3,
  collapse = TRUE,
  comment = "#>"
  )
options(digits = 3)

We provide S4 generic functions (plot and lines) for fitted model. As a result, it is easy to compare different models and graphically examine outliers through multiple dose-response curves.

Outliers at significance levels 0.01 and 0.05 are indicated by different colors and symbols. Below we show the LL.3, LL.3u and sigEmax curves in this example corresponding to the three-parameter log-logistic model with lower limit 0, three-parameter log-logistic with upper limit 1 and the sigmoid Emax model.

library(drexplorer2)
data(ryegrass)

fit.LL.3 <- drFit(drMat = ryegrass[, c(2, 1)],
                  modelName = "LL.3", alpha = 0.05, fitCtr = FALSE)
fit.LL.3u <- drFit(drMat = ryegrass[, c(2, 1)],
                   modelName = "LL.3u", alpha = 0.05, fitCtr = FALSE)
fit.sigEmax <- drFit(drMat = ryegrass[, c(2, 1)],
                     modelName = "sigEmax", alpha = 0.05, fitCtr = FALSE)

plot(fit.LL.3, main = '', col = 4, lwd = 2)
lines(fit.LL.3u, col = 5, lwd = 2)
lines(fit.sigEmax, col = 6, lwd = 2)
legend("bottomleft", c('LL.3', 'LL.3u', 'sigEmax'), col = 4:6, lwd = 3, bty = "n")

With these many models fitted, which one should be preferred? One way is to look at the Residual Standard Error (RSE) as below. We see that the LL.3u model is best by the RSE criteria.

sapply(list(fit.LL.3, fit.LL.3u, fit.sigEmax), function(x) x@info$RSE)

We also compare the curves using sigEmax model with and without outlier identification.

# no outlier excluded
fit.sigEmax0 <- drFit(drMat = ryegrass[, c(2, 1)],
                      modelName = "sigEmax", alpha = 1, fitCtr = FALSE)

plot(fit.sigEmax0, main = 'sigEmax model', col = 7, lwd = 2)
lines(fit.sigEmax, col = 6, lwd = 2)
legend("bottomleft", c('alpha = 0.05', 'ignored'), col = c(6, 7), lwd = 3, bty = "n")