denoiseq {denoiSeq}

Differential expression analysis using a bottom-up model

Description

The denoiseq function perfoms default analysis by first normalising the counts and then estimating the model parameters using Bayesian inference. Size factors are estimated from count matrix and used for the normalisation. The Gibb's sampling algorithm is then used to sample from the joint posterior distribution of the model parameters.

Usage

denoiseq(RDobject, steps, tuningSteps = floor(steps/3))

Arguments

RDobject

A readsData object.

steps

An integer representing the number of iterations.

tuningSteps

An integer representing the number of iterations to be used for tuning the step sizes. Defaulted to a third of steps.

Details

The denoiSeq package is based on a bottom-up model for PCR sequencing developed by Ndifon et al. (2012). The model generates, in a bottom-up manner, a probability distribution for the final copy number of a gene, that is a superposition of the negative binomial and the binomial distributions. The derived distribution has three main parameters, i.e N, p and f, which represent the initial gene amount before amplification, the amplification efficiency and the dilution rate, respectively.

Bayesian inference is used to estimate the model parameters. The counts in each column are used to estimate the size factors (Anders and Huber, 2010) which are in turn used to normalise the counts. For an m by n matrix, inference aims at estimating the three sets of parameters, i.e p, f and N_i ’s (2m in total because we are considering 2 conditions with the same m genes in each). denoiseq uses the rows in each condition to estimate parameter N_i for each gene in that condition, and uses the entire dataset, combined from both conditions, to estimate p and f.

For differential expression analysis, the primary parameters of interest are N_{iA} and N_{iB} (from conditions A and B respectively), for each gene i.

Value

The same readsData object but with a filled output slot. The output slot now contains 2 lists, i.e samples which contains posterior samples for each of the parameters N_i, p and f, and stepsize which contains the tuned step sizes.

Examples

#pre -filtering to remove lowly expressed genes
ERCC <- ERCC[rowSums(ERCC)>0, ]
RD <- new('readsData', counts = ERCC)
steps <- 30
#30 steps are used for illustration here. Atleast 5000 steps are adequate.
BI <- denoiseq(RD, steps)