Using purrr and modelr to make predictions from dose-response models (drm)

Question

I have a data set that looks, in abbreviate form, like this:

library(tidyverse)    
dat_s<-tibble(
      type=c(rep("A", 9), rep("B", 8), rep("C", 10)),
      ref=c("ref3", "ref3", "ref1",  "ref2", "ref2", "ref1", "ref2", "ref2", "ref2", "ref2", 
            "ref1", "ref2", "ref2", "ref3", "ref2", "ref3", "ref1", "ref3", 
            "ref2", "ref3", "ref1", "ref1", "ref3", "ref1", "ref1", "ref2", "ref2"),
      info=as.character(sample(100, 27)),
      liv=c(3.0e-05, 2.9e-07, 2.2e-07, 2.7e-07, 2.6e-06, 4.8e-07, 1.4e-05, 2.6e-06, 7.7e-06, 2.2e-06, 
            1.5e-07, 1.6e-07, 1.8e-06, 6.1e-08, 4.9e-06, 4.9e-06, 1.8e-06, 1.5e-07,
            4.3e-08, 1.8e-06, 1.0e-07, 1.6e-07, 9.7e-07, 1.0e-06, 6.4e-07, 1.2e-07, 5.7e-06),
      prod=c(0.00, 2, 3, 4.80, 2.10, 5.10, 0.00, 0.13, 2.00, 0.13, 0.00, 4.10, 4.60, 2.10, 0.26, 0.00, 
             4.60, 0.00, 4.60, 2.10, 4.80, 0.00, 0.00, 1.80, 3.60, 4.10, 0.00)
    )%>%
  mutate(livp1=liv+1)

I want to calculate dose response relationships for each combination of type and ref, make predictions to plot a curve, and calculate residuals. The info column is to reflect that I have additional columns in this data frame which I need to preserve, but are not important in the dose-response analysis.

I start by creating the models using a function and a nested data frame:

dr_s<-function(df){drc::drm(data=df, prod~log(livp1), fct=drc::LL.3())}

dat_mods<-
  dat_s%>%
  group_by(type, ref)%>%
  nest() %>%
  mutate(dr_mod=map(data, dr_s))

Which works to create the models and put them in the data frame. To use add_predictions with models of the type drm, the input has to be a data.frame (rather than a tibble). When I try to add predictions for each livp1 variable (according to the comments below):

dat_mods%>%

mutate(mod_preds=map2(data, dr_mod, ~add_predictions(data=as.data.frame(.x), model=.y))))

I get a non-numeric argument to binary operator error message. This code works fine when the info column is not character class. However, I need to retain this information with the predicted data, and would like to avoid pulling it from the data frame if possible.

Any guidance is appreciated!

Answer 1

This one was pretty silly, the predict method for models of class drm doesn't work on objects of class tibble. So you have to convert .x to data.frame.

dat_s%>%
  group_by(type, ref)%>%
  nest() %>%
  mutate(dr_mod=map(data, dr_s),
         mod_preds=map2(data, dr_mod, 
                        ~modelr::add_predictions(data = as.data.frame(.x[,"livp1"]),
                                                 model = .y)))
## A tibble: 9 x 5
## Groups:   type, ref [9]
#  type  ref   data             dr_mod mod_preds       
#  <chr> <chr> <list>           <list> <list>          
#1 A     ref3  <tibble [2 × 4]> <drc>  <df[,2] [2 × 2]>
#2 A     ref1  <tibble [2 × 4]> <drc>  <df[,2] [2 × 2]>
#3 A     ref2  <tibble [5 × 4]> <drc>  <df[,2] [5 × 2]>
#4 B     ref2  <tibble [4 × 4]> <drc>  <df[,2] [4 × 2]>
#5 B     ref1  <tibble [2 × 4]> <drc>  <df[,2] [2 × 2]>
#6 B     ref3  <tibble [2 × 4]> <drc>  <df[,2] [2 × 2]>
#7 C     ref3  <tibble [3 × 4]> <drc>  <df[,2] [3 × 2]>
#8 C     ref2  <tibble [3 × 4]> <drc>  <df[,2] [3 × 2]>
#9 C     ref1  <tibble [4 × 4]> <drc>  <df[,2] [4 × 2]>