Fit a dose-response curve

Fits a dose-response model to well-level summaries of a treatment across doses. Supports 4-parameter log-logistic ("4pl"), 3-parameter log-logistic with fixed lower asymptote at zero ("3pl"), and a simple linear model ("linear").

Usage

cr_dose_response(
  experiment,
  channel,
  treatment = NULL,
  model = c("4pl", "3pl", "linear"),
  log_dose = TRUE
)

Arguments

experiment

A cr_experiment.

channel

Channel name.

treatment

Character. Names of treatments to include (for example a compound at several doses). If NULL, the entire design is used after filtering by dose > 0.

model

"4pl", "3pl" or "linear".

log_dose

If TRUE, the fit is done on log10 doses. Doses must all be positive in that case.

Value

A list with class "cr_dose_response" containing the fitted model, the data used for the fit, the estimated parameters, and helper predictions.

Details

The 4PL model fitted is: $$y = d + \frac{a - d}{1 + (x / e)^b}$$ where a is the top asymptote, d the bottom asymptote, e the inflection point (EC50 / IC50), and b the Hill slope.

Examples

exp <- cr_example_experiment(seed = 1, n_cells_per_well = 30)
# Add a synthetic dose-response sub-design
exp$design$dose <- ifelse(exp$design$treatment == "CompoundA_high",
                          500, exp$design$dose)
fit <- cr_dose_response(exp, channel = "marker_1", model = "4pl")
print(fit$params)
#> # A tibble: 2 × 3
#>   parameter estimate std_error
#>   <chr>        <dbl>     <dbl>
#> 1 intercept   -3887.        NA
#> 2 slope        3217.        NA