Fit a Bayesian Sigmoid Model — sigmoid

Fits the sigmoid model. The functional form is

<response> ~ sigmoid(ac50, hill, top, bottom, <treatment>)

where

sigmoid = bottom + (top - bottom)/(1 + 10^((ac50 - <treatment>) * hill))

By default the observed data (and therefore should be columns in the input data data.frame) are

treatment: log_dose, the log10 of the dose as a molar concentration
response: response, with unspecified units

and the modeled parameters are

ac50: the dose where the response reaches half maximal activity
hill: the hill coefficient controlling the slope at the ac50, by convention the slope of an agonist is positive
top: the response when => Inf
bottom: the response when => -Inf

To configure the model you can use the following helper functions for agonist models

sigmoid_agonist_formula(): define how the response relates to the treatment and the parameters with ac50->ec50, and how any covariates predict the parameters
sigmoid_agonist_prior(): define the prior for the model parameters, with the slope constrained to be positive
sigmoid_agonist_init(): define initial values for the model parameters to match the default sigmoid_agonist_prior()

the following helper functions for an antagonist model

sigmoid_antagonist_formula(): define how the response relates to the treatment and the parameters with ac50->ic50, and how any covariates predict the parameters
sigmoid_antagonist_prior(): define the prior for the model parameters, with the slope being constrained be negative
sigmoid_antagonist_init(): define initial values for the model parameters to match the default sigmoid_antagonist_prior()

Usage

sigmoid_model(
  data,
  formula = sigmoid_agonist_formula(),
  prior = sigmoid_agonist_prior(),
  init = sigmoid_agonist_init(),
  iter = 8000,
  control = list(adapt_delta = 0.99),
  stanvar_function = sigmoid_stanvar(),
  expose_functions = TRUE,
  ...
)

Arguments

data: data.frame of observed data. It must contain columns for the treatment, response and any additional predictors specified in the formula. See sigmoid_agonist_formula() for more details.
formula: bpformula formula for the sigmoid model (see above).
prior: brmsprior defining a distribution for the model parameters defined in the formula. (see above).
init: function to initialize brms::brm model for for the parameters (see above)
iter: numeric of iterations the model runs. Increasing iter can help with model convergence (see brms::brm)
control: a named list of arguments to control the sampler's behavior. Adding max_treedepth and giving a greater value than 10 can improve model convergence (see brms::brm)
stanvar_function: stan code for the model.
expose_functions: logical. Expose the sigmoid function used in the model. This is needed e.g. for brms::loo_compare()
...: additional arguments passed to brms::brm

Value

bpfit object, which is a wrapper around a brms::brmsfit object.

Examples

if (FALSE) { # \dontrun{
  BayesPharma::sigmoid_model(
    data = data,
    formula = BayesPharma::sigmoid_antagonist_formula(predictors = 0 + drug),
    prior = BayesPharma::sigmoid_antagonist_prior(),
    init = BayesPharma::sigmoid_antagonist_init())
} # }