R/swan2.R

In wateRmelon: Illumina 450 methylation array normalization and metrics

# code from minfi minimally modified to not require minfi objects

swan <- function (mn, un, qc, da=NULL, return.MethylSet=FALSE ) {

#LS#  make a MethylSet containing the data and an RGChannelSet just for the annotation
#    if (is.null(mSet)) 
#        mSet <- preprocessRaw(rgSet)
#   if(!library(IlluminaHumanMethylation450kmanifest, logical.return=TRUE, quietly=TRUE)){
#         stop('can\'t load IlluminaHumanMethylation450kmanifest package')
#    }

    mSet   <-  new("MethylSet", Meth = mn, Unmeth = un)  
    rgSet  <- new('RGChannelSet', annotation="IlluminaHumanMethylation450k")


    if (!is.null(da)){
       IlluminaHumanMethylation450kmanifest@data$TypeI$nCpG<-
     da[IlluminaHumanMethylation450kmanifest@data$TypeI$Name,'nCpG']  
    } 

#LS#
    
  typeI <- getProbeInfo(rgSet, type = "I")[, c("Name", "nCpG")]
   typeII <- getProbeInfo(rgSet, type = "II")[, c("Name", "nCpG")]
   CpG.counts <- rbind(data.frame(typeI@listData), data.frame(typeII@listData))
   CpG.counts$Name <- as.character(CpG.counts$Name)
   CpG.counts$Type <- rep(c("I", "II"), times = c(nrow(typeI), 
       nrow(typeII)))
   names(CpG.counts)[2] <- "CpGs"
   counts <- CpG.counts[CpG.counts$Name %in% featureNames(mSet), 
       ]
   subset <- min(table(counts$CpGs[counts$Type == "I" & counts$CpGs
%in% 
       1:3]), table(counts$CpGs[counts$Type == "II" & counts$CpGs %in% 
       1:3]))

# see email from Jovana Maksimovic to Leo 29 oct 2012

#LS#
    bg <- bgIntensitySwan.methylumi(qc)
    #LS#

    methData <- getMeth(mSet)
    unmethData <- getUnmeth(mSet)
    normMethData <- NULL
    normUnmethData <- NULL
    xNormSet <- vector("list", 2)
    xNormSet[[1]] <- minfi:::getSubset(counts$CpGs[counts$Type == "I"], 
        subset)
    xNormSet[[2]] <- minfi:::getSubset(counts$CpGs[counts$Type == "II"], 
        subset)
    for (i in 1:ncol(mSet)) {
        message(sprintf("Normalizing array %d of %d\n", i, ncol(mSet)))
        normMethData <- cbind(normMethData, minfi:::normaliseChannel(methData[rownames(methData) %in% 
            counts$Name[counts$Type == "I"], i], methData[rownames(methData) %in% 
            counts$Name[counts$Type == "II"], i], xNormSet, bg[i]))
        normUnmethData <- cbind(normUnmethData, minfi:::normaliseChannel(unmethData[rownames(unmethData) %in% 
            counts$Name[counts$Type == "I"], i], unmethData[rownames(unmethData) %in% 
            counts$Name[counts$Type == "II"], i], xNormSet, bg[i]))
    }
    colnames(normMethData) <- sampleNames(mSet)
    colnames(normUnmethData) <- sampleNames(mSet)
    normSet <- mSet
    assayDataElement(normSet, "Meth") <- normMethData
    assayDataElement(normSet, "Unmeth") <- normUnmethData
    normSet@preprocessMethod <- c(sprintf("SWAN (based on a MethylSet preprocesses as '%s'", 
        mSet@preprocessMethod[1]), as.character(packageVersion("minfi")), 
        as.character(packageVersion("IlluminaHumanMethylation450kmanifest")))
    if(return.MethylSet) return(normSet)
    getBeta(normSet)
}




bgIntensitySwan.methylumi  <- function (rg) { # rg is a list of 
                                              # red and green matrices
    # rg <- intensitiesByChannel(QCdata(methylumi_obj))
    neg <- grep("NEGATIVE",rownames(rg$Cy3)) 

    grnMed <- colMedians(rg$Cy3[neg, ])
    redMed <- colMedians(rg$Cy5[neg, ])
    return(rowMeans(cbind(grnMed, redMed)))
}