Fit_Single_Hill: Estimate parameters in Hill model families

In gilles-guillot/MDR: Test Loewe additivity with Model Deviation Ratio

Description

Estimate parameters in single-compound dose-reponse models under some of the Hill model families (2,3,4 parameters)

Parameters

Usage

Fit_Single_Hill(
  x,
  y,
  n,
  SCDR_model = c("Hill2", "Hill3", "Hill4"),
  lower = c(1e-04, 1e-04, 1e-04, 1e-04),
  upper = c(10, 10, 1, 1),
  bootstrap = FALSE,
  B = 100,
  select = FALSE,
  eps = 1e-15,
  verbose = TRUE
)

Arguments

x	sequence of dose input
y	sequence of responses
n	number of binomial samples at various dose values
SCDR_model	one of 2-4 parameter Hill function
lower	lower limits for the search of parameter
upper	upper limits for the search of parameter
bootstrap	logical resampling for estimating parameter variability
B	number of bootstrap resampling
select	TRUE/FALSE perform model selection
eps	used to replace 0's by small numerical values
verbose	will print the percentage of computation carried out if TRUE

Value

Estimate of Hill model(s) parameters and bootstrap sample for this estimate Perform model selection by cross-validation Return estimate of 50 NB: 50

Detail Model assumed: y ~ B(n,f(x)) Hill models: Hill 4: f(x) = d + (c-d) / ( 1+ (a/x)**b) Hill 3: d=0 Hill 2: d=0 ; c=1 ML estimation, optimization of log likelihood with Generalized Simulated Annealing (GenSA::GenSA)

0 doses bring numerical issues 0's in x and y replaced by min(x) * 1e-9 / min(y) * 1e-9

Model selection with leave-"some"-out likelihood cross-validation "some" = all obs receiving a certain dose level

Author(s)

Gilles Guillot [e-mail]

gilles-guillot/MDR documentation built on Jan. 21, 2020, 8:09 a.m.