duplicateDiscordanceAcrossDatasets: Functions to check discordance and allelic dosage correlation across datasets

Description

These functions compare genotypes in pairs of duplicate scans of the same sample across multiple datasets. 'duplicateDiscordanceAcrossDatasets' finds the number of discordant genotypes both by scan and by SNP. 'dupDosageCorAcrossDatasets' calculates correlations between allelic dosages both by scan and by SNP, allowing for comparision between imputed datasets or between imputed and observed - i.e., where one or more of the datasets contains continuous dosage [0,2] rather than discrete allele counts {0,1,2}.

Usage

duplicateDiscordanceAcrossDatasets(genoData1, genoData2, match.snps.on=c("position", "alleles"), subjName.cols, snpName.cols=NULL, one.pair.per.subj=TRUE, minor.allele.only=FALSE, missing.fail=c(FALSE, FALSE), scan.exclude1=NULL, scan.exclude2=NULL, snp.exclude1=NULL, snp.exclude2=NULL, snp.include=NULL, verbose=TRUE)
minorAlleleDetectionAccuracy(genoData1, genoData2, match.snps.on=c("position", "alleles"), subjName.cols, snpName.cols=NULL, missing.fail=TRUE, scan.exclude1=NULL, scan.exclude2=NULL, snp.exclude1=NULL, snp.exclude2=NULL, snp.include=NULL, verbose=TRUE)
dupDosageCorAcrossDatasets(genoData1, genoData2, match.snps.on=c("position", "alleles"), subjName.cols="subjectID", snpName.cols=NULL, scan.exclude1=NULL, scan.exclude2=NULL, snp.exclude1=NULL, snp.exclude2=NULL, snp.include=NULL, snp.block.size=5000, scan.block.size=100, verbose=TRUE)

Arguments

genoData1

GenotypeData object containing the first dataset.

genoData2

GenotypeData object containing the second dataset.

match.snps.on

One or more of ("position", "alleles", "name") indicating how to match SNPs. "position" will match SNPs on chromosome and position, "alleles" will also require the same alleles (but A/B designations need not be the same), and "name" will match on the columns give in snpName.cols.

subjName.cols

2-element character vector indicating the names of the annotation variables that will be identical for duplicate scans in the two datasets. (Alternatively, one character value that will be recycled).

snpName.cols

2-element character vector indicating the names of the annotation variables that will be identical for the same SNPs in the two datasets. (Alternatively, one character value that will be recycled).

one.pair.per.subj

A logical indicating whether a single pair of scans should be randomly selected for each subject with more than 2 scans.

minor.allele.only

A logical indicating whether discordance should be calculated only between pairs of scans in which at least one scan has a genotype with the minor allele (i.e., exclude major allele homozygotes).

missing.fail

For duplicateDiscordanceAcrossDatasets, a 2-element logical vector indicating whether missing values in datasets 1 and 2, respectively, will be considered failures (discordances with called genotypes in the other dataset). For minorAlleleDetectionAccuracy, a single logical indicating whether missing values in dataset 2 will be considered false negatives (missing.fail=TRUE) or ignored (missing.fail=FALSE).

scan.exclude1

An integer vector containing the ids of scans to be excluded from the first dataset.

scan.exclude2

An integer vector containing the ids of scans to be excluded from the second dataset.

snp.exclude1

An integer vector containing the ids of SNPs to be excluded from the first dataset.

snp.exclude2

An integer vector containing the ids of SNPs to be excluded from the second dataset.

snp.include

List of SNPs to include in the comparison. Should match the contents of the columns referred to by snpName.cols. Only valid if match.snps.on includes "name".

snp.block.size

Block size for SNPs

scan.block.size

Block size for scans

verbose

Logical value specifying whether to show progress information.

Value

chromosome: chromosome
position: base pair position
snpID1: snpID from genoData1
snpID2: snpID from genoData2
alleles: alleles sorted alphabetically
alleleA1: allele A from genoData1
alleleB1: allele B from genoData2
alleleA2: allele A from genoData2
alleleB2: allele B from genoData2
name: the common SNP name given in snpName.cols
discordant: number of discordant pairs
npair: number of pairs examined
n.disc.subj: number of subjects with at least one discordance
discord.rate: discordance rate i.e. discordant/npair
npair: number of sample pairs compared (non-missing in genoData1)
sensitivity: sensitivity
specificity: specificity
positivePredictiveValue: Positive predictive value
negativePredictiveValue: Negative predictive value
nsamp.dosageR: number of samples in r calculation (i.e., non missing data in both genoData1 and genoData2)
dosageR: dosage correlation
subjectID: subject-level identifier for duplicate sample pair
scanID1: scanID from genoData1
scanID2: scanID from genoData2
nsnp.dosageR: number of SNPs in r calculation (i.e., non missing data in both genoData1 and genoData2)
dosageR: dosage correlation

Details

duplicateDiscordanceAcrossDatasets calculates discordance metrics both by scan and by SNP. If one.pair.per.subj=TRUE (the default), each subject with more than two duplicate genotyping instances will have one scan from each dataset randomly selected for computing discordance. If one.pair.per.subj=FALSE, discordances will be calculated pair-wise for all possible cross-dataset pairs for each subject.

dupDosageCorAcrossDatasets calculates dosage correlation (Pearson correlation coefficient) both by scan and by SNP. Note it only allows for one pair of duplicate scans per sample. For this function only, genoData1 and genoData2 must have been created with GdsGenotypeReader objects.

By default, overlapping variants are identified based on position and alleles. Alleles are determined via 'getAlleleA' and 'getAlleleB' accessors, so users should ensure these variables are referring to the same strand orientation in both datests (e.g., both plus strand alleles). It is not necessary for the A/B ordering to be consistent across datasets. For example, two variants at the same position with alleleA="C" and alleleB="T" in genoData1 and alleleA="T" and alleleB="C" in genoData2 will stil be identified as overlapping.

If minor.allele.only=TRUE, the allele frequency will be calculated in genoData1, using only samples common to both datasets.

If snp.include=NULL (the default), discordances will be found for all SNPs common to both datasets.

genoData1 and genoData2 should each have "alleleA" and "alleleB" defined in their SNP annotation. If allele coding cannot be found, the two datasets are assumed to have identical coding. Note that 'dupDosageCorAcrossDatasets' can NOT detect where strand-ambiguous (A/T or C/G) SNPs are annotated on different strands, although the r2 in these instances would be unaffected: r may be negative but r2 will be positive.

minorAlleleDetectionAccuracy summarizes the accuracy of minor allele detection in genoData2 with respect to genoData1 (the "gold standard"). TP=number of true positives, TN=number of true negatives, FP=number of false positives, and FN=number of false negatives. Accuracy is represented by four metrics:

sensitivity for each SNP as TP/(TP+FN)
specificity for each SNP as TN/(TN+FP)
positive predictive value for each SNP as TP/(TP+FP)
negative predictive value for each SNP as TN/(TN+FN).

TP, TN, FP, and FN are calculated as follows:

			genoData1
			mm
Mm	MM		mm
2TP	1TP + 1FP	2FP	genoData2
Mm	1TP + 1FN	1TN + 1TP	1TN + 1FP
	MM	2FN	1FN + 1TN
2TN		--	2FN
1FN

"M" is the major allele and "m" is the minor allele (as calculated in genoData1). "-" is a missing call in genoData2. Missing calls in genoData1 are ignored. If missing.fail=FALSE, missing calls in genoData2 (the last row of the table) are also ignored.

Examples

Run this code

# first set
snp1 <- data.frame(snpID=1:10, chromosome=1L, position=101:110,
                   rsID=paste("rs", 101:110, sep=""),
                   alleleA="A", alleleB="G", stringsAsFactors=FALSE)
scan1 <- data.frame(scanID=1:3, subjectID=c("A","B","C"), sex="F", stringsAsFactors=FALSE)
mgr <- MatrixGenotypeReader(genotype=matrix(c(0,1,2), ncol=3, nrow=10), snpID=snp1$snpID,
                            chromosome=snp1$chromosome, position=snp1$position, scanID=1:3)
genoData1 <- GenotypeData(mgr, snpAnnot=SnpAnnotationDataFrame(snp1),
                          scanAnnot=ScanAnnotationDataFrame(scan1))

# second set
snp2 <- data.frame(snpID=1:5, chromosome=1L,
                   position=as.integer(c(101,103,105,107,107)),
                   rsID=c("rs101", "rs103", "rs105", "rs107", "rsXXX"),
                   alleleA= c("A","C","G","A","A"),
                   alleleB=c("G","T","A","G","G"),
                   stringsAsFactors=FALSE)
scan2 <- data.frame(scanID=1:3, subjectID=c("A","C","C"), sex="F", stringsAsFactors=FALSE)
mgr <- MatrixGenotypeReader(genotype=matrix(c(1,2,0), ncol=3, nrow=5), snpID=snp2$snpID,
                            chromosome=snp2$chromosome, position=snp2$position, scanID=1:3)
genoData2 <- GenotypeData(mgr, snpAnnot=SnpAnnotationDataFrame(snp2),
                          scanAnnot=ScanAnnotationDataFrame(scan2))

duplicateDiscordanceAcrossDatasets(genoData1, genoData2,
  match.snps.on="position",
  subjName.cols="subjectID")

duplicateDiscordanceAcrossDatasets(genoData1, genoData2,
  match.snps.on=c("position", "alleles"),
  subjName.cols="subjectID")

duplicateDiscordanceAcrossDatasets(genoData1, genoData2,
  match.snps.on=c("position", "alleles", "name"),
  subjName.cols="subjectID",
  snpName.cols="rsID")

duplicateDiscordanceAcrossDatasets(genoData1, genoData2,
  subjName.cols="subjectID",
  one.pair.per.subj=FALSE)

minorAlleleDetectionAccuracy(genoData1, genoData2,
  subjName.cols="subjectID")

dupDosageCorAcrossDatasets(genoData1, genoData2,
  scan.exclude2=scan2$scanID[duplicated(scan2$subjectID)])

Run the code above in your browser using DataLab