mcmc.qpcr.classic {MCMC.qpcr}

Analyzes qRT-PCR data using "classic" model, based on multigene normalization.

Description

Normalizes the data using specified control genes, fits a single model to estimate changes at all genes. Use for datasets with not too many Cq values above 30.

Usage

mcmc.qpcr.classic(fixed = NULL, globalFixed = NULL, random = NULL, globalRandom = NULL,
data, controls, genebysample = TRUE, geneSpecRes=FALSE, center = TRUE, ...)

Arguments

fixed

desired combination of fixed effects, as a text string. Do not use "*" symbol, list it fully, such as: 'factor1+factor2+factor1:factor2'.

globalFixed

Vector of fixed covariates (categorical or continuous) that are expected to affect all genes in the sample in the same way. These would be typically related to quality and/or quantity of RNA, such as RIN value.

random

A vector of names for gene-specific scalar random effects, such as 'c("effect1","effect2")'.

globalRandom

Random covariates (categorical only) affecting all genes, similar to globalFixed.

data

output of the cq2log() function

controls

Vector of control gene names. These will be pushed to the back of the gene list during model fitting, in the reverse order.

genebysample

Whether random gene by sample interactions should be modeled as an additional random effect. If the model fails to converge, specify 'genebysample=F'.

geneSpecRes

Whether the model should include gene-specific residuals. This was the default in MCMC.qpcr v.1.0; now it is switched off since it does not have any visible effect on the results (as long as genebysample=TRUE), and only makes the model converge slower or fail to converge.

center

Whether to center the normalized log-transformed Cq values arond the mean for each gene. Centering does not affect the inference; it only makes the plots of the results more comprehensible (in my opinion).

...

other options for MCMCglmm function, such as nitt (number of iterations), thin (tinning interval), and burnin (number of initial iterations to disregard). For a more precise inference (but longer runs) specify 'nitt=45000, thin=20, burnin=5000'. See MCMCglmm documentation for more details.

Details

This function takes an as input the log-transformed relative abundance values, performs multigene normalization as per Vandesompele et al 2002, and runs an MCMC chain to fit a lognormal linear mixed model to estimate gene expression changes jointly at all genes. It is very powerful, as long as the average stability of control genes can be trusted.

This function requires data prepared by cq2log function and must have the control genes specified; otherwise the arguments and syntax are similar to the mcmc.qpcr function.

Value

An MCMCglmm object. See mcmc.qpcr function for details and examples.

Author(s)

Mikhail V. Matz, University of Texas at Austin

References

Matz MV, Wright RM, Scott JG (2013) No Control Genes Required: Bayesian Analysis of qRT-PCR Data. PLoS ONE 8(8): e71448. doi:10.1371/journal.pone.0071448