hlaPredMerge: Merge prediction results from multiple HIBAG models
In zhengxwen/HIBAG: HLA Genotype Imputation with Attribute Bagging

hlaPredMerge

R Documentation

Merge prediction results from multiple HIBAG models

Description

Return an object of hlaAlleleClass, which contains predicted HLA types.

Usage

hlaPredMerge(..., weight=NULL, equivalence=NULL, use.matching=TRUE,
    ret.dosage=TRUE, ret.postprob=FALSE, max.resolution="", rm.suffix=FALSE,
    verbose=TRUE)

Arguments

`...`	The object(s) of `hlaAlleleClass`, having a field of 'postprob', and returned by `hlaPredict(..., type="response+prob")`
`weight`	the weight used for each prediction; if `NULL`, equal weights to be used; or set the weight vector to be the training sample sizes
`equivalence`	a `data.frame` with two columns, the first column for new equivalent alleles, and the second for the alleles possibly exist in the object(s) passed to this function; there is no replace if the allele is not found in the second column
`use.matching`	if `TRUE`, use actual probabilities (i.e., poster prob. * matching) for merging; otherwise, use poster prob. instead. `use.matching=TRUE` is recommended.
`ret.dosage`	if `TRUE`, return dosages
`ret.postprob`	if `TRUE`, return average posterior probabilities
`max.resolution`	"2-digit", "1-field", "4-digit", "2-field", "6-digit", "3-field", "8-digit", "4-field", "allele", "protein", "full", "none", or "": "allele" = "2-digit"; "protein" = "4-digit"; "full", "none" or "" for no limit on resolution
`rm.suffix`	whether remove the non-digit suffix in the last field, e.g., for "01:22N", "N" is a non-digit suffix
`verbose`	if `TRUE`, show information

Details

Calculate a new probability matrix for each pair of HLA alleles, by averaging (posterior) probabilities from all models with specified weights. If equivalence is specified, multiple alleles might be collapsed into one class.

Value

Return a hlaAlleleClass object.

Author(s)

Xiuwen Zheng

Examples

# make a "hlaAlleleClass" object
hla.id <- "A"
hla <- hlaAllele(HLA_Type_Table$sample.id,
    H1 = HLA_Type_Table[, paste(hla.id, ".1", sep="")],
    H2 = HLA_Type_Table[, paste(hla.id, ".2", sep="")],
    locus=hla.id, assembly="hg19")

# divide HLA types randomly
set.seed(100)
hlatab <- hlaSplitAllele(hla, train.prop=0.5)
names(hlatab)
# "training"   "validation"
summary(hlatab$training)
summary(hlatab$validation)

# SNP predictors within the flanking region on each side
region <- 500   # kb
snpid <- hlaFlankingSNP(HapMap_CEU_Geno$snp.id, HapMap_CEU_Geno$snp.position,
    hla.id, region*1000, assembly="hg19")
length(snpid)  # 275

# training and validation genotypes
train.geno <- hlaGenoSubset(HapMap_CEU_Geno,
    snp.sel=match(snpid, HapMap_CEU_Geno$snp.id),
    samp.sel=match(hlatab$training$value$sample.id,
    HapMap_CEU_Geno$sample.id))
test.geno <- hlaGenoSubset(HapMap_CEU_Geno,
    samp.sel=match(hlatab$validation$value$sample.id,
    HapMap_CEU_Geno$sample.id))

# train HIBAG models
set.seed(100)

# please use "nclassifier=100" when you use HIBAG for real data
m1 <- hlaAttrBagging(hlatab$training, train.geno, nclassifier=2,
    verbose.detail=TRUE)
m2 <- hlaAttrBagging(hlatab$training, train.geno, nclassifier=2,
    verbose.detail=TRUE)


# validation
pd1 <- hlaPredict(m1, test.geno, type="response+prob")
pd2 <- hlaPredict(m2, test.geno, type="response+prob")

hlaCompareAllele(hlatab$validation, pd1)$overall
hlaCompareAllele(hlatab$validation, pd2)$overall

# merge predictions from multiple models, by voting from all classifiers
pd <- hlaPredMerge(pd1, pd2)
pd

hlaCompareAllele(hlatab$validation, pd)$overall

# collapse to 2-digit
pd <- hlaPredMerge(pd1, pd2, max.resolution="2-digit", ret.postprob=FALSE)
pd

zhengxwen/HIBAG documentation built on Nov. 19, 2024, 1:01 p.m.