ibs: Computes (average) Idenity-by-State for a set of people and... In GenABEL: genome-wide SNP association analysis

Description

Given a set of SNPs, computes a matrix of average IBS for a group of individuals. This function facilitates quality control of genomic data. E.g. people with exteremly high (close to 1) IBS may indicate duplicated samples (or twins), simply high values of IBS may indicate relatives.

Usage

 1 2  ibs(data, snpsubset, idsubset = NULL, cross.idsubset = NULL, weight = "no", snpfreq = NULL) 

Arguments

 data object of snp.data-class snpsubset Index, character or logical vector with subset of SNPs to run analysis on. If missing, all SNPs from data are used for analysis. idsubset IDs of people to be analysed. If missing, all people from data are used for analysis. cross.idsubset Parameter allowing parallel implementation. Not to be used normally. If supplied together with idsubset, the ibs/kinship for all pairs between idsubset and cross.idsubset computed. weight "no" for direct IBS computations, "freq" to weight by allelic frequency asuming HWE and "eVar" for empirical variance to be used snpfreq when option weight="freq" used, you can provide fixed allele frequencies

Details

When weight "freq" is used, IBS for a pair of people i and j is computed as

f_{i,j} = \frac{1}{N} Σ_k \frac{(x_{i,k} - p_k) * (x_{j,k} - p_k)}{(p_k * (1 - p_k))}

where k changes from 1 to N = number of SNPs GW, x_{i,k} is a genotype of ith person at the kth SNP, coded as 0, 1/2, 1 and p_k is the frequency of the "+" allele. This apparently provides an unbiased estimate of the kinship coefficient.

With "eVar" option above formula changes by using ( 2 * empirical variance of the genotype ) in the denominator. The empirical variance is computed according to the formula

Var(g_k) = \frac{1}{M} Σ_i g_{ik}^2 - E[g_k]^2

where M is the number of people

Only with "freq" option monomorphic SNPs are regarded as non-informative.

ibs() operation may be very lengthy for a large number of people.

Value

A (Npeople X Npeople) matrix giving average IBS (kinship) values between a pair below the diagonal and number of SNP genotype measured for both members of the pair above the diagonal.

On the diagonal, homozygosity 0.5*(1+inbreeding) is provided with option 'freq'; with option 'eVar' the diagonal is set to 0.5; the diagonal is set to homozygosity with option 'no'.

attr(computedobject,"Var") returns variance (replacing the diagonal when the object is used by egscore

Author(s)

Yurii Aulchenko

check.marker, summary.snp.data, snp.data-class
  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 require(GenABEL.data) data(ge03d2c) set.seed(7) # compute IBS based on a random sample of 1000 autosomal marker selectedSnps <- sample(autosomal(ge03d2c),1000,replace=FALSE) a <- ibs(ge03d2c,snps=selectedSnps) a[1:5,1:5] mds <- cmdscale(as.dist(1-a)) plot(mds) # identify smaller cluster of outliers km <- kmeans(mds,centers=2,nstart=1000) cl1 <- names(which(km$cluster==1)) cl2 <- names(which(km$cluster==2)) if (length(cl1) > length(cl2)) cl1 <- cl2; cl1 # PAINT THE OUTLIERS IN RED points(mds[cl1,],pch=19,col="red") # compute genomic kinship matrix to be used with e.g. polygenic, mmscore, etc a <- ibs(ge03d2c,snps=selectedSnps,weight="freq") a[1:5,1:5] # now replace diagonal with EIGENSTRAT-type of diaganal to be used for egscore diag(a) <- hom(ge03d2c[,autosomal(ge03d2c)])\$Var a[1:5,1:5] ############################## # compare 'freq' with 'eVar' ############################## ibsFreq <- ibs(ge03d2c,snps=selectedSnps, weight="freq") ibsEvar <- ibs(ge03d2c,snps=selectedSnps, weight="eVar") mdsEvar <- cmdscale( as.dist( 0.5 - ibsEvar ) ) plot(mdsEvar) outliers <- (mdsEvar[,1]>0.1) ibsFreq[upper.tri(ibsFreq,diag=TRUE)] <- NA ibsEvar[upper.tri(ibsEvar,diag=TRUE)] <- NA plot(ibsEvar,ibsFreq) points(ibsEvar[outliers,outliers],ibsFreq[outliers,outliers],col="red")