aroma-project.org

Author: Henrik Bengtsson
Created on: 2010-05-17
Last updated on: 2012-02-02

The CRMAv2 (Bengtsson et al. 2009) is the recommended method for estimating full-resolution copy numbers (CN) from all Affymetrix SNP and CN chip types, including custom-made ones.   CRMAv2 can be done either by doing each step explicitly as in the vignettes, or using the following doCRMAv2() and doASCRMAv2() "block" methods.

Usage:

ds <- doCRMAv2("HapMap270,testSet", chipType="GenomeWideSNP_6");
ds <- doCRMAv2("HapMap270,testSet", chipType="GenomeWideSNP_6,Full"); 
ds <- doCRMAv2("HapMap270,testSet", chipType="GenomeWideSNP_5");
ds <- doCRMAv2("HapMap270,testSet", chipType="GenomeWideSNP_5,Full,r2"); 
ds <- doCRMAv2("HapMap270,testSet", chipType="Mapping250K_Nsp", plm="RmaCnPlm");
ds <- doCRMAv2("HapMap270,testSet", chipType="Mapping50K_Hind240", plm="RmaCnPlm");
ds <- doCRMAv2("GSE8605", chipType="Mapping10K_Xba142", plm="RmaCnPlm");

Argument 'verbose': Since CRMAv2 will take minutes or hours, we suggest that you add argument 'verbose=-10' to see some verbose output while processing the data set.

Argument 'plm': Note how we specify plm="RmaCnPlm" for the 10K-500K platforms.  We do this in order to control for probe-affinity effects, which are possible to estimate for those chip types.  Probe-affinity effects are not possible to estimate in the newer chip types (e.g. GWS) because there all replicated probes are technical/identical replicates.  See the CRMAv2 paper for more details.

Single- vs multi-array method:  When fitting the CRMAv2 model with plm="RmaCnPlm", the method is a multi-array method, whereas with the default (plm="AvgCnPlm"), then method is a single-array method.  In the latter case the results for one array will be independent of the other arrays in the same data set.  This make is possible to process a subsets of the arrays and postponing the others for later and still get the same results.  See the CRMAv2 paper for more details.

Process a subset of the arrays:  In case the single-array method (plm="AvgCnPlm") is used it is possible to process a subset of the arrays without having to worry about what arrays to including in the batch.  The results will be the same regardless.  For instance, the following two cases will give identical results:

dsA <- doCRMAv2("HapMap270,testSet", chipType="GenomeWideSNP_6");
dsA <- extract(dsA, arrays=c(6,1,3:4));

dsB <- doCRMAv2("HapMap270,testSet", chipType="GenomeWideSNP_6", arrays=c(6,1,3:4));

Note, this is only the case if plm="AvgCnPlm".

Allele-specific copy-number (ASCN) estimates:  By adding argument combineAlleles=FALSE to the above, allele-specific CNs will be estimated (for SNPs).  The default is to estimate total CNs (combineAlleles=TRUE).  Alternatively, use doASCRMAv2(), which is short for doCRMAv2(..., combineAlleles=FALSE).  The estimated total CNs will differ slightly when using the two alternatives and the CRMAv2 method has been optimized for total CNs (as in CRMAv2 paper).  However, if you plan to do ASCN analysis, we recommend to use combineAlleles=FALSE from the beginning, because the statistical performance is almost as good while you will save lots of time only processing the data once.