R: Computing point estimate for the model parameters for all probe sets.

RPA.pointestimate {RPA}

R Documentation

Computing point estimate for the model parameters for all probe sets.

Description

Computes point estimate

Usage

RPA.pointestimate(abatch, sets = NULL, myseed = 101, 
                         priors = NULL, 
                         epsilon = 10^(-2), cind = FALSE, 
                         sigma2.method = "var",
                         d.method = "fast",
                         verbose = TRUE)

Arguments

`abatch`	An AffyBatch object.
`sets`	Specifies the probesets for which RPA estimates will be computed. Default: all probe sets.
`myseed`	Specifies the random seed.
`priors`	An 'rpa.priors' object. Can be used to set user-specified priors for the model parameters. Used only with sigma2.method = "basic".
`epsilon`	Convergence tolerance. The iteration is deemed converged when the change in all parameters is < epsilon.
`cind`	Specifies which of the arrays in abatch is used as a control in computing probe-level differential expression.
`sigma2.method`	Optimization method for sigma2 (probe-specific variances). "basic": optimization using user-specified alpha, beta priors. Default: alpha, beta = 1e-6. "var": utilizes the fact that the cost function converges to variance with large sample sizes. Default method.
`d.method`	Method to optimize d. "basic": Finds a mode by directly optimizing the cost function of the probabilistic model. "fast": weighted mean over the probes, weighted by inverse probe-specific variances. The solution converges to this with large sample size. Default method.
`verbose`	Print progress information during computation. Default: TRUE.

Details

Assuming data set S with P observations of signal d with Gaussian noise that is specific for each observation (specified by a vector sigma2 of length P), this method gives a point estimate of d and sigma2. Note that probe-level variance priors alpha, beta are used only when sigma2.method = "basic". The sigma2.method = "var" assumes non-informative priors. The d.method = "fast" is the preferred method for point computing point estimates when sample size is large. It computes the average over probe-level observations, weighted by the inverse probe-specific variances, and is expected to be more robust and faster than d.method="basic" that finds point estimate for d by directly optimizing the posterior distribution.

Value

An instance of the 'rpa' class. This is an extended list containing the following elements:

`d`	A matrix of probesets x arrays. Specifies the estimated 'true' underlying differential gene expression signal over the arrays (vs. the control array 'cind') for each investigated probeset. Note that the control array is not included.
`sigma2`	A list. Each element corresponds to a probeset, and contains a vector that gives the estimated variance for each probe in that probeset.
`cind`	Specifies which of the arrays in the abatch (the affybatch object to be analyzed) was used as a control in computing probe-level differential expression.
`sets`	A character vector listing the investigated probesets.

Note

With large sample size, sigma2.method="var" and d.method="fast" are recommended. With small sample size and informative prior, sigma2.method="basic" and d.method="basic" may be preferable.

Author(s)

Leo Lahti <leo.lahti@tkk.fi>

References

Probabilistic Analysis of Probe Reliability in Differential Gene Expression Studies with Short Oligonucleotide Arrays. Lahti et al., TCBB/IEEE, to appear. See http://www.cis.hut.fi/projects/mi/software/RPA/

Examples


# Load example data set
require(affydata)
data(Dilution)

## Run RPA analysis     
## Compute RPA for the whole data set   
## Slow, not executed here      
##rpa.results <- RPA.pointestimate(Dilution)
        
# Compute RPA for specific probesets only       
sets <- geneNames(Dilution)[1:5]        
rpa.results <- RPA.pointestimate(Dilution,sets)

# Visualize the results for one of the probe sets
plot(rpa.results[sets[[1]],])

[Package RPA version 1.1.2 Index]