invClust: 'invClust' Class and Methods

In isglobal-brge/invClust: inference of invarsion alleles from biallelic haplotype clustering

Description

Mixture model fitting with Hardy Weinberg Equilibrium and population stratification to infer haplotype (inversion) alleles.

Usage

invClust(roi, wh = 1, geno, annot, SNPtagg="n", SNPsel=1:ncol(geno), method=1, dim = 1, pc = 0, ngroups = 1, ...)

Arguments

roi	text file of data.frame with Region of Interest information. Four columns are required: chr, LBP, RBP, reg, with chromosome, left break point, right break point and a character "reg" that identifies the inversion
wh	which ROI (row of roi) to be considered for computation.
geno	genotypes in snpMatrix format of snpStats
annot	snp annotation as .map PLINK format, or read.plink()$map from snpStats. The columns "chromosome", "names", "position" are required and have to conform to this naming.
SNPtagg	set SNPtagg="y" to use tagg SNPs in roi for tagging the haplotype groups
SNPsel	vector with snps to be selected for computation
method	method=1 performs EM algorithm for three genotypes, method=2, performs a clustering within the the genotypes for an additional third haplotype
dim	either 1 or 2 indicating the number fo MDS compoents to be used
pc	if population clustering is to be performed, first component of a genome-wide PCA of geno
ngroups	maximum number of subpopualtions to be considered
...	control arguments for the EM algorithm: it (it=1000) maximum number fo iterations, tol (tol=10e-5) convergence tolerance

Details

invClust computes the biallelic haplotypes in Hardy Weinberg Equilibrium (with the possibility of clustering by geographical subpopulations) that may underlie an inversion event. It fits a mixture model with an expectation maximization routine, only controled by convergence criteria. Initial conditions are general for a wide range of cases. Clustering is performed in 1 or 2 dimensions of mutidimeansional scaling (argument dim) and, if geographical subpopulation is considered, the first component of a genome wide PCA (argument pc). In this last case, a visualization of the PCA analysis can inform on the suitable number of groups to be considered (argument ngroup).

Each subject in the sample is assigned a probability to a given genotype (NN, NI, II) which can be recovered by x["genotypes"], where x is of class invClust ( e.g. the result of an invClust call). Most probable genotypes can be extracted with getGenotypes(x). In a similar way, if subpopulation classification is considered, probability for group membership is recovered with x["groups"].

Plots are also implemented for this class, plot(x) will display the clustered data on the fitted distribution, according to dimensions used. For inclussion of subpopulation classification, selection of the marginals can be done though a plot argument wh=c("yy","xy"). wh="like" plots the likelihood with respect to the number of itereations only for dim=1.

A useful quantity is a quality score (getQuality(x)) that computes the overlap integral of the cluster components, a value of 1 gives no overlap while 0 refers to complete overlap.

Value

EMestimate	List with fitted parameters
datin	List with data used to fit the model: x for firt PCA component and y MDS components

Author(s)

Alejandro Caceres

Examples

data(geno)
inv<-invClust(roi=roi,wh=1,geno=geno,annot=annot,dim=1)
plot(inv)
head(inv["genotypes"])

isglobal-brge/invClust documentation built on May 19, 2020, 5:19 a.m.