qpcR (version 1.3-7.1)

pcrsim: Simulation of sigmoidal qPCR data with goodness-of-fit analysis

Description

Simulated sigmoidal qPCR curves are generated from an initial model to which some user-defined homoscedastic or heteroscedastic noise is added. One or more models can then be fit to this random data and goodness-of-fit (GOF) measures are calculated for each of the models. This is essentially a Monte-Carlo approach testing for the best model in dependence to some noise structure in sigmodal models.

Usage

pcrsim(object, nsim = 100, error = 0.02, 
       errfun = function(y) 1, plot = TRUE,
       fitmodel = NULL, select = FALSE, 
       statfun = function(y) mean(y, na.rm = TRUE),
       PRESS = FALSE, ...)

Arguments

object

an object of class 'pcrfit.

nsim

the number of simulated curves.

error

the gaussian error used for the simulation. See 'Details'.

errfun

an optional function for the error distribution. See 'Details'.

plot

should the simulated and fitted curves be displayed?

fitmodel

a model or model list to test against the initial model.

select

if TRUE, a matrix is returned with the best model in respect to each of the GOF measures.

statfun

a function to be finally applied to all collected GOF measures, default is the average.

PRESS

logical. If set to TRUE, the computationally expensive PRESS statistic will be calculated.

...

other parameters to be passed on to plot or pcrfit.

Value

• A list containing the following items:
• cyc
  same as in 'arguments'.
• fluoMat
  a matrix with the simulated qPCR data in columns.
• coefList
  a list with the coefficients from the fits for each model, as subitems.
• gofList
  a list with the GOF measures for each model, as subitems.
• statList
  a list with the GOF measures summarized by statfun for each model, as subitems.
• modelMat
  if select = TRUE, a matrix with the best model for each GOF measure and each simulation.

Details

The value defined under error is just the standard deviation added plainly to each y value from the initial model, thus generating a dataset with homoscedastic error. With aid of errfun, the distribution of the error along the y values can be altered and be used to generate heteroscedastic error along the curve, i.e. as a function of the magnitude. Example: errfun = function(y) 1 same variance for all y, as is. errfun = function(y) y variance as a function of the y-magnitude. errfun = function(y) 1/y variance as an inverse function of the y-magnitude. For the effect, see 'Examples'.

Examples

## generate initial model
m1 <- pcrfit(reps, 1, 2, l4)

## simulate homoscedastic error
## and test l4 and l5 on data
res1 <- pcrsim(m1, error = 0.2, nsim = 20, 
               fitmodel = list(l4, l5))

## use heteroscedastic noise typical for 
## qPCR: more noise at lower fluorescence
res2 <- pcrsim(m1, error = 0.01, errfun = function(y) 1/y,
              nsim = 20, fitmodel = list(l4, l5, l6))

## get 95% confidence interval for 
## the models GOF in question (l4, l5, l6) 
res3 <- pcrsim(m1, error = 0.2, nsim = 20, fitmodel = list(l4, l5, l6),
              statfun = function(y) quantile(y, c(0.025, 0.975)))
res3$statList  

## count the selection of the 'true' model (l4)
## for each of the GOF measures,
## use PRESS statistic => SLOW!
## BIC wins!!
res4 <- pcrsim(m1, error = 0.05, nsim = 20, fitmodel = list(l3, l4, l5),
               select = TRUE, PRESS = TRUE)
apply(res4$modelMat, 2, function(x) sum(x == 2))