@@ -1,7 +1,7 @@

 Package: crlmm

 Type: Package

 Title: Genotype Calling (CRLMM) and Copy Number Analysis tool for Affymetrix SNP 5.0 and 6.0 and Illumina arrays.

-Version: 1.9.4

+Version: 1.9.5

 Date: 2010-10-17

 Author: Benilton S Carvalho <carvalho@bclab.org>, Robert Scharpf <rscharpf@jhsph.edu>, Matt Ritchie <mritchie@wehi.edu.au>, Ingo Ruczinski <iruczins@jhsph.edu>, Rafael A Irizarry

 Maintainer: Benilton S Carvalho <carvalho@bclab.org>, Robert Scharpf <rscharpf@jhsph.edu>, Matt Ritchie <mritchie@wehi.EDU.AU>

@@ -1,6 +1,6 @@

 useDynLib("crlmm", .registration=TRUE)

 ## this is temporary

-exportPattern("^[^\\.]")

+## exportPattern("^[^\\.]")

 ##---------------------------------------------------------------------------

 ## Biobase

@@ -906,21 +906,25 @@ fit.lm4 <- function(strata,

 		}

 		tmp <- cbind(medianA.AA.snp[, k], medianB.BB.snp[,k], phiA.snp[, k], phiB.snp[, k])

 		tmp <- tmp[rowSums(is.na(tmp) == 0) & rowSums(tmp < 20) == 0, ]

-		stopifnot(nrow(tmp) > 100)

+		if(nrow(tmp) < 100){

+			stop("too few markers for estimating nonpolymorphic CN on chromosome X")

+		}

 		X <- cbind(1, log2(c(tmp[, 1], tmp[, 2])))

 		Y <- log2(c(tmp[, 3], tmp[, 4]))

 		betahat <- solve(crossprod(X), crossprod(X, Y))

 		if(enough.men){

-			X.men <- cbind(1, medianA.AA.M[, k])

+			##X.men <- cbind(1, medianA.AA.M[, k])

+			X.men <- cbind(1, log2(medianA.AA.M[, k]))

 			Yhat1 <- as.numeric(X.men %*% betahat)

 			## put intercept and slope for men in nuB and phiB

 			phiB[, k] <- 2^(Yhat1)

-			nuB[, k] <- 2^(medianA.AA.M[, k]) - phiB[, k]

+			##nuB[, k] <- 2^(medianA.AA.M[, k]) - phiB[, k]

+			nuB[, k] <- medianA.AA.M[, k] - 1*phiB[, k] ## male has 1 copy

 		}

-		X.fem <- cbind(1, medianA.AA.F[, k])

+		X.fem <- cbind(1, log2(medianA.AA.F[, k]))

 		Yhat2 <- as.numeric(X.fem %*% betahat)

 		phiA[, k] <- 2^(Yhat2)

-		nuA[, k] <- 2^(medianA.AA.F[, k]) - 2*phiA[, k]

+		nuA[, k] <- medianA.AA.F[, k] - 2*phiA[, k]

 	}

 	if(THR.NU.PHI){

 		nuA[nuA < MIN.NU] <- MIN.NU

@@ -1685,6 +1689,7 @@ summarizeSnps <- function(strata,

 	CP <- as.matrix(snpCallProbability(object)[index, ])

 	AA <- as.matrix(A(object)[index, ])

 	BB <- as.matrix(B(object)[index, ])

+	FL <- as.matrix(flags(object)[index, ])

 	if(verbose) message("        Computing summaries...")

 	for(k in seq_along(batches)){

 		B <- batches[[k]]

@@ -1695,6 +1700,12 @@ summarizeSnps <- function(strata,

 		xx <- CP[, B]

 		highConf <- (1-exp(-xx/1000)) > GT.CONF.THR

 		G <- G*highConf

+		## Some markers may have genotype confidences scores that are ALL below the threshold

+		## For these markers, provide statistical summaries based on all the samples and flag

+		## Provide summaries for these markers and flag to indicate the problem

+		ii <- which(rowSums(G) == 0)

+		G[ii, ] <- GG[ii, B]

+		## table(rowSums(G==0))

 		##G <- G*highConf*NORM

 		A <- AA[, B]

 		B <- BB[, B]

@@ -1771,7 +1782,7 @@ crlmmCopynumber <- function(object,

 			    prior.prob=rep(1/4,4),

 			    seed=1,

 			    verbose=TRUE,

-			    GT.CONF.THR=0.95,

+			    GT.CONF.THR=0.80,

 			    MIN.NU=2^3,

 			    MIN.PHI=2^3,

 			    THR.NU.PHI=TRUE,

@@ -1797,14 +1808,20 @@ crlmmCopynumber <- function(object,

 		       X.SNP="chromosome X SNPs",

 		       X.NP="chromosome X nonpolymorphic markers")

 	}

-	if(verbose) message("Computing summary statistics of the genotype clusters for each batch")

+	if(verbose & is.lds) {

+		message("Large data support with ff package is enabled.")

+		message("To reduce RAM, the markers are divided into several strata (see ?ocProbesets for details) and summary statistics are computed on each stratum")

+	}

 	for(i in seq_along(type)){

 		## do all types

 		marker.type <- type[i]

 		if(verbose) message(paste("...", mylabel(marker.type)))

 		##if(verbose) message(paste("Computing summary statistics for ", mylabel(marker.type), " genotype clusters for each batch")

-		marker.index <- whichMarkers(marker.type, is.snp,

-					     is.autosome, is.annotated, is.X)

+		marker.index <- whichMarkers(marker.type,

+					     is.snp,

+					     is.autosome,

+					     is.annotated,

+					     is.X)

 		object <- genotypeSummary(object=object,

 					  GT.CONF.THR=GT.CONF.THR,

 					  type=marker.type,

@@ -456,28 +456,28 @@ ACN <- function(object, allele, i , j){

 				marker.index <- i[is.X & !is.snp]

 				acn.index <- which(is.X & !is.snp)

 				acn[acn.index, ] <- NA

-##				female.index <- j[object$gender[j] == 2]

-##				## 3. CHR X NPs: women

-##				if(length(female.index) > 0){

-##					female.batch.index <- match(unique(as.character(batch(object))[female.index]), batchNames(object))

-##					jj <- which(object$gender[j] == 2)

-##					acn[acn.index, jj] <- C1(object, marker.index, female.batch.index, female.index)

-##				}

+				female.index <- j[object$gender[j] == 2]

+				## 3. CHR X NPs: women

+				if(length(female.index) > 0){

+					female.batch.index <- match(unique(as.character(batch(object))[female.index]), batchNames(object))

+					jj <- which(object$gender[j] == 2)

+					acn[acn.index, jj] <- C1(object, marker.index, female.batch.index, female.index)

+				}

 ##				## 4. CHR X NPs: men

-##				male.index <- j[object$gender[j] == 1]

-##				if(length(male.index) > 0){

-##					if(is.ff){

-##						open(nuB(object))

-##						open(phiB(object))

-##					}

-##					male.batch.index <- match(unique(as.character(batch(object))[male.index]), batchNames(object))

-##					jj <- which(object$gender[j] == 1)

-##					acn[acn.index, jj] <- C2(object, marker.index, male.batch.index, male.index, NP.X=TRUE)

-##					if(is.ff){

-##						close(nuB(object))

-##						close(phiB(object))

-##					}

-##				}

+				male.index <- j[object$gender[j] == 1]

+				if(length(male.index) > 0){

+					if(is.ff){

+						open(nuB(object))

+						open(phiB(object))

+					}

+					male.batch.index <- match(unique(as.character(batch(object))[male.index]), batchNames(object))

+					jj <- which(object$gender[j] == 1)

+					acn[acn.index, jj] <- C2(object, marker.index, male.batch.index, male.index, NP.X=TRUE)

+					if(is.ff){

+						close(nuB(object))

+						close(phiB(object))

+					}

+				}

 			}

 		}

 		if(is.ff){

@@ -51,19 +51,19 @@ platforms must have a corresponding annotation package (installed

 separately) that are listed below.

 <<annotationPackages>>=

-annotationPackages()

+pkgs <- annotationPackages()

+pkgs <- pgks[grep("Crlmm", pkgs)]

+pkgs

 @

-Only the annotation package that end with the 'Crlmm' postfix are

-supported for copy number estimation. For instance,

-'pd.genomewidesnp6' and 'genomewidesnp6Crlmm' are both annotation

-packages for the Affymetrix 6.0 platform, but the

-'genomewidesnp6Crlmm' annotation package must be used for copy number

-estimation.  The annotation package is specified through the 'cdfName'

-we specify the cdf name for Affymetrix 6.0, provide the complete path

-to the CEL files, and indicate where intermediate files from the copy

-number estimation are to be saved.

+Note that 'pd.genomewidesnp6' and 'genomewidesnp6Crlmm' are both

+annotation packages for the Affymetrix 6.0 platform available on

+Bioconductor, but the 'genomewidesnp6Crlmm' annotation package must be

+used for copy number estimation.  The annotation package is specified

+through the 'cdfName' -- the identifier without the 'Crlmm' postfix.

+In the following code, we specify the cdf name for Affymetrix 6.0,

+provide the complete path to the CEL files, and indicate where

+intermediate files from the copy number estimation are to be saved.

 <<setup>>=

 cdfName <- "genomewidesnp6"

@@ -188,10 +188,11 @@ The \Rfunction{crlmmCopynumber} performs the following steps:

   displayed during the processing.

 <<LDS_copynumber>>=

-cnSet <- checkExists("cnSet", .path=outdir, .FUN=crlmmCopynumber, object=gtSet)

+GT.CONF.THR <- 0.90

+cnSet <- checkExists("cnSet", .path=outdir, .FUN=crlmmCopynumber, object=gtSet, GT.CONF.THR=GT.CONF.THR)

 @

-In an effort to reduct I/O, the \Rpackage{crlmmCopynumber} function no

+In an effort to reduce I/O, the \Rpackage{crlmmCopynumber} function no

 longer stores the allele-specific estimates of copy number as part of

 the object.  Rather, several functions are available that will compute

 relatively quickly the allele-specific estimates from the stored

@@ -476,6 +477,118 @@ Scatterplots of the A and B allele intensities (log-scale) can be

 useful for assessing the biallelic genotype calls.  This section of

 the vignette is currently under development.

+\section{Reasons for missing values}

+

+There are several reasons for estimates of the allele-specific copy

+number to have missing values (\texttt{NA}'s). This section briefly

+elaborates on the source of missing values in the HapMap analysis and

+discusses possible alternatives to reduce the number of missing

+values.  Note that allele-specific copy number, 'CA' and 'CB', is not

+saved in the \Robject{cnSet} object.  Rather, the respective accessors

+calculate 'CA' and 'CB' on the fly from the normalized intensities and

+from the marker-specific parameter estimates in the linear model.  In

+general, a missing value arises when the background or slope parameter

+was not estimated in the linear model. Most often, missing values

+occur when the genotype confidence scores for a SNP were below the

+threshold used by the \Robject{crlmmCopynumber} function. For the

+HapMap analysis, we used a confidence threshold of

+\Sexpr{GT.CONF.THR}.

+

+<<NA>>=

+autosome.index <- which(isSnp(cnSet) & chromosome(cnSet) < 23)

+sample.index <- which(batch(cnSet) == "C")

+ca <- CA(cnSet, i=autosome.index, j=sample.index)

+missing.ca <- is.na(ca)

+(nmissing.ca <- sum(missing.ca))

+@

+

+If \Robject{nmissing.ca} is nonzero, then one could check the genotype

+confidence scores provided by the crlmm genotyping algorithm against

+the threshold specified in \Robject{crlmmCopynumber}.

+

+<<NAconfidenceScores>>=

+if(nmissing.ca > 0){

+	gt.conf <- i2p(snpCallProbability(cnSet)[autosome.index, sample.index])

+	below.thr <- gt.conf < GT.CONF.THR

+	index.allbelow <- as.integer(which(rowSums(below.thr) == length(sample.index)))

+	all.equal(index, index.allbelow)

+	## or calculate the proportion of missing effected by low crlmm confidence

+	sum(index.allbelow %in% index)/length(index)

+}

+@

+

+We repeat the above check for missing values at polymorphic loci on

+chromosome X.  In this case, we compare the \Robject{rowSums} of the

+missing values to the number of samples to check whether all of the

+estimates are missing for a given SNP.

+

+<<NAchromosomeX>>=

+X.index <- which(isSnp(cnSet) & chromosome(cnSet) == 23)

+ca.X <- CA(cnSet, i=X.index, j=sample.index)

+missing.caX <- is.na(ca.X)

+nmissing.caX <- sum(missing.caX)

+missing.snp.index <- which(rowSums(missing.caX) == length(sample.index))

+index <- which(rowSums(missing.caX) == length(sample.index))

+length(index)*length(sample.index)/nmissing.caX

+@

+

+From the above codechunk, we see that \Sexpr{length(index)} SNPs have

+NAs for all the samples. Next, we tally the number of NAs for

+polymorphic markers on chromosome X that are below the confidence

+threshold.  For the HapMap analysis, all of the missing values arose

+from SNPs in which either the men or the women had confidence scores

+that were all below the threshold.

+

+<<NAcrlmmConfidenceScore>>=

+if(nmissing.caX > 0){

+	gt.conf <- i2p(snpCallProbability(cnSet)[X.index, sample.index])

+	below.thr <- gt.conf < GT.CONF.THR

+	F <- which(cnSet$gender[sample.index] == 2)

+	M <- which(cnSet$gender[sample.index] == 1)

+	index.allbelowF <- as.integer(which(rowSums(below.thr[, F]) == length(F)))

+	index.allbelowM <- as.integer(which(rowSums(below.thr[, M]) == length(M)))

+	index.allbelow <- as.integer(unique(c(index.allbelowF, index.allbelowM)))

+	all.equal(index, index.allbelow)

+	## or calculate the proportion of missing effected by low crlmm confidence

+	sum(index.allbelow %in% index)/length(index)

+}

+@

+

+Next, we verify that the number of missing values is the same for the

+'B' allele at autosomal polymorphic loci

+

+<<NAmissingForBallele>>=

+cb <- CB(cnSet, i=autosome.index, j=sample.index)

+missing.cb <- is.na(cb)

+nmissing.cb <- sum(missing.cb)

+@

+

+For nonpolymorphic loci, the genotype confidence score is irrelevant

+and estimates should be available at all markers.

+

+<<NAnonpolymorphic.autosome>>=

+np.index <- which(!isSnp(cnSet) & chromosome(cnSet)==23)

+cb <- CB(cnSet, i=np.index, j=sample.index)

+stopifnot(sum(is.na(cb)) == 0)

+ca.F <- CA(cnSet, i=np.index, j=F)

+ca.M <- CA(cnSet, i=np.index, j=M)

+sum(is.na(ca.F))

+sum(is.na(ca.M))

+@

+

+In total, there were \Sexpr{length(missing.snp.index)} polymorphic

+markers on chromosome X for which copy number estimates are not

+available.  Lowering the confidence threshold would permit estimation

+of copy number at most of these loci.  A confidence threshold is

+included as a parameter for the copy number estimation as an approach

+to reduce the sensitivity of genotype-specific summary statistics,

+such as the within-genotype median, to intensities from samples that

+do not clearly fall into one of the biallelic genotype clusters. There

+are drawbacks to this approach, including variance estimates that are

+overly optimistic.  More direct approaches for outlier detection and

+removal may be explored in the future.

+

+

 \section{Session information}

 <<sessionInfo, results=tex>>=

 toLatex(sessionInfo())

@@ -10,7 +10,7 @@

 crlmmCopynumber(object, MIN.SAMPLES=10, SNRMin = 5, MIN.OBS = 1, 

 	        DF.PRIOR = 50, bias.adj = FALSE,

                 prior.prob = rep(1/4, 4), seed = 1, verbose = TRUE,

-                GT.CONF.THR = 0.95, MIN.NU = 2^3, MIN.PHI = 2^3,

+                GT.CONF.THR = 0.80, MIN.NU = 2^3, MIN.PHI = 2^3,

                 THR.NU.PHI = TRUE, type=c("SNP", "NP", "X.SNP", "X.NP"))

 }