hapassoc (version 0.2)

PreEM: Pre-process the data before fitting it with EM

Description

This function takes as an argument the original dataframe with non-SNP and SNP data and converts the genotype data at single SNPs (the single-locus genotypes) into haplotype data. The rows of the original data frame should correspond to subjects and each SNP should have two columns, one for each allele of the single-locus genotype. The SNP data should comprise the last 2*numSNPs columns. If the haplotypes for a subject cannot be inferred from his or her genotype data, "pseudo-individuals" representing all possible haplotype combinations consistent with the single-locus genotypes are considered. Missing single-locus genotypes, up to a maximum of maxMissingGenos (see below), are allowed, but subjects with missing data in more than maxMissingGenos, or with missing non-SNP data, are removed. Initial estimates of haplotype frequencies are then obtained using the EM algorithm applied to the multilocus genotype data. Haplotypes with frequencies below a user-specified tolerance (zero.tol) are assumed not to exist and are removed from further consideration. (Pseudo-individuals having haplotypes of negligible frequency are deleted and the column in the design matrix corresponding to that haplotype is deleted.) For the remaining haplotypes, those with non-negligible frequency below a user-defined pooling tolerance (pooling.tol) are pooled into a single category called "pooled" in the design matrix for the risk model. However, the frequencies of each of these pooled haplotypes are still calculated separately.

Usage

PreEM(dat,numSNPs,maxMissingGenos=1,pooling.tol = 0.05, zero.tol = 1/(2 * nrow(dat) * 10)

Arguments

dat

the non-SNP and SNP data as a data frame. The SNP data should comprise the last 2*numSNPs columns.

numSNPs

number of SNPs per haplotype

maxMissingGenos

maximum number of single-locus genotypes with missing data to allow for each subject. (Subjects with more missing data, or with missing non-SNP data are removed.) The default is 1.

pooling.tol

pooling tolerance -- by default set to 0.05

zero.tol

tolerance for haplotype frequencies below which haplotypes are assumed not to exist -- by default set to $\frac{1}{2*N*10}$ where N is the number of subjects

Value

• haplotest
  T/F, true if some haplotypes were pooled in the risk model
• initGamma
  initial estimates of haplotype frequencies
• zeroFreqHaplos
  list of haplos assumed not to exist
• pooledHaplos
  list of haplos pooled into a single category in the design matrix
• nonHaploDM
  non-haplotype portion of the AUGMENTED data frame
• haploDM
  data frame with $2^{numSNPs}$ columns scoring number of copies of each haplotype for each pseudo-individual
• haploMat
  matrix with 2 columns giving haplotypes for each pseudo-individual
• wt
  vector giving initial weights for each pseudo-individual for the EM algorithm
• ID
  index for each individual in the original data frame. Note that all pseudo-individuals have the same ID value
• unknown
  vector indicating whether the haplotype information was missing for each row in the augmented data

See Also

EM,summary.EM.

Examples

data(hypoDat)
example.preEM<-PreEM(hypoDat, numSNPs=3)

# To get the initial haplotype frequencies:
example.preEM$initGamma
#      h000       h001       h010       h011       h100       h101       h110 
#0.25179111 0.26050418 0.23606001 0.09164470 0.10133627 0.02636844 0.01081260 
#      h111 
#0.02148268 
# The '001' haplotype is estimated to be the most frequent

example.preEM$pooledHaplos
# "h101" "h110" "h111"
# These haplotypes are to be pooled in the design matrix for the risk model

names(example.preEM$haploDM)
# "h000"   "h001"   "h010"   "h011"   "h100"   "pooled"