Using `dose` in a model causes it to have a Symengine parse error

[billdenney]

It would be better if dose were allowed as a variable in a model. Changing dose to dosex made this work. Is there a reason we cannot use dose?

library(nlmixr2)

min_df <- data.frame(
  ID = as.factor(c(1, 1, 2, 2)),
  AMT = c(10, 20, 10, 20),
  DV = c(1, 2, 3, 4),                            # Number of successes
  total_trials = as.integer(c(5, 5, 6, 6))  # Number of trials
)
min_df_expanded <-
  min_df |>
  rename(
    DV_orig = DV,
    # AMT is a reserved column name for the dose to be administered
    dose = AMT
  ) |>
  # make each success and failure one row
  mutate(
    dosex = dose,
    DV =
      pmap(
        .l = list(success = DV_orig, total = total_trials),
        .f = function(success, total) c(rep(1, success), rep(0, total - success))
      )
  ) |>
  unnest(cols = DV) |>
  mutate(
    # must have TIME and CMT even when it is meaningless
    TIME = 1,
    cmt = "arbitrary"
  )

min_model_dose <- function() {
  ini({
    t_bot <- 0.1
    t_top <- 0.9
    t_ic50 <- 10.0
    t_hill <- 1.5
    size_est <- 5
    eta.bot ~ 0.1
    eta.top ~ 0.1
    eta.ic50 ~ 0.1
    eta.hill ~ 0.1
  })
  model({
    bot <- exp(t_bot + eta.bot)
    top <- exp(t_top + eta.top)
    ic50 <- exp(t_ic50 + eta.ic50)
    hill <- exp(t_hill + eta.hill)
    prob <- bot + (top - bot) / (1 + (dose / ic50)^hill)
    DV ~ dbinom(size_est, prob)
  })
}

min_model_dosex <- function() {
  ini({
    t_bot <- 0.1
    t_top <- 0.9
    t_ic50 <- 10.0
    t_hill <- 1.5
    size_est <- 5
    eta.bot ~ 0.1
    eta.top ~ 0.1
    eta.ic50 ~ 0.1
    eta.hill ~ 0.1
  })
  model({
    bot <- exp(t_bot + eta.bot)
    top <- exp(t_top + eta.top)
    ic50 <- exp(t_ic50 + eta.ic50)
    hill <- exp(t_hill + eta.hill)
    prob <- bot + (top - bot) / (1 + (dosex / ic50)^hill)
    DV ~ dbinom(size_est, prob)
  })
}

nlmixr2(min_model_dose,
        data = min_df_expanded,
        est = "focei"
        )
#> ℹ parameter labels from comments are typically ignored in non-interactive mode
#> ℹ Need to run with the source intact to parse comments
#> → loading into symengine environment...
#> → pruning branches (`if`/`else`) of full model...
#> ✔ done
#> Error in s4vecbasic_mut_append(ans, x) : SymEngine exception: Parse error
#> Error: SymEngine exception: Parse error
nlmixr2(min_model_dosex,
        data = min_df_expanded,
        est = "focei"
)
#> ℹ parameter labels from comments are typically ignored in non-interactive mode
# Edit: removed a bunch of lines
#> → compress parHistData in nlmixr2 object, save 9400

^{Created on 2025-09-29 with reprex v2.1.1}

[mattfidler]

Nope. I will not support dose in the model since it is a function dose() and dose(cmt). The parsing to get dose as a function and a covariate is difficult at best, especially in something that parses into symengine.

That being said it should have a better error.

[mattfidler]

Well, perhaps it can be done, if I already support it in low-level rxode2, will have to see later. But this is not a high priority for me right now.

Also need to add dose() to the function table; it seems to be missing.

[billdenney]

A clear error would be a big benefit. (I understand that it is not a general high priority.)