anomIdentifyLowQuality: Identify low quality samples
In GWASTools: Tools for Genome Wide Association Studies

Description Usage Arguments Details Value Author(s) See Also Examples

Identify low quality samples for which false positive rate for anomaly detection is likely to be high. Measures of noise (high variance) and high segmentation are used.

1 2	anomIdentifyLowQuality(snp.annot, med.sd, seg.info, sd.thresh, sng.seg.thresh, auto.seg.thresh)

`snp.annot`	`SnpAnnotationDataFrame` with column "eligible", where "eligible" is a logical vector indicating whether a SNP is eligible for consideration in anomaly detection (usually FALSE for HLA and XTR regions, failed SNPs, and intensity-only SNPs). See `HLA` and `pseudoautosomal`.
`med.sd`	data.frame of median standard deviation of BAlleleFrequency (BAF) or LogRRatio (LRR) values across autosomes for each scan, with columns "scanID" and "med.sd". Usually the result of `medianSdOverAutosomes`. Usually only eligible SNPs are used in these computations. In addition, for BAF, homozygous SNPS are excluded.
`seg.info`	data.frame with segmentation information from `anomDetectBAF` or `anomDetectLOH`. Columns must include "scanID", "chromosome", and "num.segs". (For `anomDetectBAF`, segmentation information is found in $seg.info from output. For `anomDetectLOH`, segmentation information is found in $base.info from output.)
`sd.thresh`	Threshold for `med.sd` above which scan is identified as low quality. Suggested values are 0.1 for BAF and 0.25 for LOH.
`sng.seg.thresh`	Threshold for segmentation factor for a given chromosome, above which the chromosome is said to be highly segmented. See Details. Suggested values are 0.0008 for BAF and 0.0048 for LOH.
`auto.seg.thresh`	Threshold for segmentation factor across autosome, above which the scan is said to be highly segmented. See Details. Suggested values are 0.0001 for BAF and 0.0006 for LOH.

Low quality samples are determined separately with regard to each of the two methods of segmentation, anomDetectBAF and anomDetectLOH. BAF anomalies (respectively LOH anomalies) found for samples identified as low quality for BAF (respectively LOH) tend to have a high false positive rate.

A scan is identified as low quality due to high variance (noise), i.e. if med.sd is above a certain threshold sd.thresh.

High segmentation is often an indication of artifactual patterns in the B Allele Frequency (BAF) or Log R Ratio values (LRR) that are not always captured by high variance. Here segmentation information is determined by anomDetectBAF or anomDetectLOH which use circular binary segmentation implemented by the R-package DNAcopy. The measure for high segmentation is a "segmentation factor" = (number of segments)/(number of eligible SNPS). A single chromosome segmentation factor uses information for one chromosome. A segmentation factor across autosomes uses the total number of segments and eligible SNPs across all autosomes. See med.sd, sd.thresh, sng.seg.thresh, and auto.seg.thresh.

A data.frame with the following columns:

`scanID`	integer id for the scan
`chrX.num.segs`	number of segments for chromosome X
`chrX.fac`	segmentation factor for chromosome X
`max.autosome`	autosome with highest single segmentation factor
`max.auto.fac`	segmentation factor for chromosome = `max.autosome`
`max.auto.num.segs`	number of segments for chromosome = `max.autosome`
`num.ch.segd`	number of chromosomes segmented, i.e. for which change points were found
`fac.all.auto`	segmentation factor across all autosomes
`med.sd`	median standard deviation of BAF (or LRR values) across autosomes. See `med.sd` in Arguments section.
`type`	one of the following, indicating reason for identification as low quality: `auto.seg`: segmentation factor `fac.all.auto` above `auto.seg.thresh` but `med.sd` acceptable `sd`: standard deviation factor `med.sd` above `sd.thresh` but `fac.all.auto` acceptable `both.sd.seg`: both high variance and high segmentation factors, `fac.all.auto` and `med.sd`, are above respective thresholds `sng.seg`: segmentation factor `max.auto.fac` is above `sng.seg.thresh` but other measures acceptable `sng.seg.X`: segmentation factor `chrX.fac` is above `sng.seg.thresh` but other measures acceptable

Cecelia Laurie

findBAFvariance, anomDetectBAF, anomDetectLOH

library(GWASdata)
data(illuminaScanADF, illuminaSnpADF)

blfile <- system.file("extdata", "illumina_bl.gds", package="GWASdata")
bl <- GdsIntensityReader(blfile)
blData <-  IntensityData(bl, scanAnnot=illuminaScanADF, snpAnnot=illuminaSnpADF)

genofile <- system.file("extdata", "illumina_geno.gds", package="GWASdata")
geno <- GdsGenotypeReader(genofile)
genoData <-  GenotypeData(geno, scanAnnot=illuminaScanADF, snpAnnot=illuminaSnpADF)

# initial scan for low quality with median SD
baf.sd <- sdByScanChromWindow(blData, genoData)
med.baf.sd <- medianSdOverAutosomes(baf.sd)
low.qual.ids <- med.baf.sd$scanID[med.baf.sd$med.sd > 0.05]

# segment and filter BAF
scan.ids <- illuminaScanADF$scanID[1:2]
chrom.ids <- unique(illuminaSnpADF$chromosome)
snp.ids <- illuminaSnpADF$snpID[illuminaSnpADF$missing.n1 < 1]
data(centromeres.hg18)
anom <- anomDetectBAF(blData, genoData, scan.ids=scan.ids, chrom.ids=chrom.ids,
  snp.ids=snp.ids, centromere=centromeres.hg18, low.qual.ids=low.qual.ids)

# further screen for low quality scans
snp.annot <- illuminaSnpADF
snp.annot$eligible <- snp.annot$missing.n1 < 1
low.qual <- anomIdentifyLowQuality(snp.annot, med.baf.sd, anom$seg.info,
  sd.thresh=0.1, sng.seg.thresh=0.0008, auto.seg.thresh=0.0001)

close(blData)
close(genoData)