calculateIncrementalRMAError: Calculate IncrementalRMA Error

In steveneschrich/IncrementalRMA: Affymetrix-based RMA Processing Using Fixed Parameters

Description

Calculate the differences between incrementalRMA and canonical RMA.

Usage

calculateIncrementalRMAError(exprs, abatch, params)

Arguments

exprs	Matrix of expression values from incrementalRMA.
abatch	A AffyBatch-class of new samples that were used for incrementalRMA.
params	An incremental parameter list as generated by parameterizeRMA.

Details

IncrementalRMA applies existing parameters to one or more new samples, without using these samples in the parameter estimation. As a result, the expression values produced will be slightly different than using RMA on all samples together. This is because the parameter estimates, including these new samples, will be slightly different. Normally, we would expect this type of difference to be small if the samples are similar to the input or if a sufficiently large training set was used to normalize samples. For instance, both frma and refRMA use large public repositories for calculating parameters. In these cases, adding one new sample should not really change the quantile normalization or row effects in median polish.

In the case of small sample sizes used for estimating parameters, a single sample could have more of an impact. However, the purpose of this package is to support precision of gene expression estimates over potential robustness. Earlier work indicated that tissue specificity could play a role in estimation as can processing technique. Therefore, the goal is to produce estimates that are as precise for the given sample set.

Given the above discussion, it would be helpful to estimate how different incrementalRMA is from just using RMA against a sample set. If the samples that were used to produce the original estimates are available (in the params) then we can compare the incrementalRMA estimates for the new samples against calculating RMA against the reference set and the new samples. Small differences suggest that the incremental process did not significantly impact the overall results, while providing the needed stability in expression estimates.

This function calculates the probeset-level error that occurred when applying incrementalRMA vs. using RMA against the full dataset. For convenience, we will consider this to be the standard error (se.exprs).

Value

A matrix of probeset-level errors between incrementalRMA and canonical RMA.

Note

Since this routine reruns RMA against a potentially large set of data, it can be quite computationally demanding.

Examples

## Not run: 
calculateIncrementalRMAError(exprs,  AffyBatch, params=list(referenceCELFiles=()))

## End(Not run)

steveneschrich/IncrementalRMA documentation built on Dec. 23, 2021, 5:32 a.m.