podkat (version 1.4.2)

computeKernel: Compute Kernel Matrix

Description

Computes kernel matrix for a given genotype matrix

Usage

computeKernel(Z, kernel=c("linear.podkat", "quadratic.podkat",
              "localsim.podkat", "linear.SKAT", "quadratic.SKAT",
              "localsim.SKAT"), weights=NULL, pos=NULL, width=1000)

Arguments

Z

a matrix or an object of class Matrix (note that the latter also includes objects of class GenotypeMatrix)

kernel

type of kernel to use

weights

numeric vector with variant weights; must be as long as the number of columns of Z. Use NULL for unweighted kernels.

pos

numeric vector with positions of variants; must be as long as the number of columns of Z. This argument is mandatory for the position-dependent kernels linear.podkat, quadratic.podkat, and localsim.podkat; ignored for kernels linear.SKAT, quadratic.SKAT, and localsim.SKAT.

width

tolerance radius parameter for position-dependent kernels linear.podkat, quadratic.podkat, and localsim.podkat (see details below); must be single positive numeric value. Ignored for kernels linear.SKAT, quadratic.SKAT, and localsim.SKAT.

Value

• a positive semi-definite kernel matrix with as many rows and columns as Z has rows

Details

This function computes a kernel matrix for a given genotype matrix Z and a given kernel. It supposes that Z is a matrix-like object (a numeric matrix, a sparse matrix, or an object of class GenotypeMatrix) in which rows correspond to samples and columns correspond to variants. There are six different kernels available: linear.podkat, quadratic.podkat, localsim.podkat, linear.SKAT, quadratic.SKAT, and localsim.SKAT. All of these kernels can be used with or without weights. The weights can be specified with the weights argument which must be a numeric vector with as many elements as the matrix Z has columns. If no weighting should be used, weights must be set to NULL.

The position-dependent kernels linear.podkat, quadratic.podkat, and localsim.podkat require the positions of the variants in Z. So, if any of these three kernels is selected, the argument pos is mandatory and must be a numeric vector with as many elements as the matrix Z has columns.

If the pos argument is NULL and Z is a GenotypeMatrix object, the positions in variantInfo(Z) are taken. In this case, all variants need to reside on the same chromosome. If the variants in variantInfo(Z) are from multiple chromosomes, computeKernel quits with an error. As said, this only happens if pos is NULL, otherwise the pos argument has priority over the information stored in variantInfo(Z).

For details on how the kernels compute the pairwise similarities of genotypes, see Subsection 9.2 of the package vignette.

References

http://www.bioinf.jku.at/software/podkat

Wu, M. C., Lee, S., Cai, T., Li, Y., Boehnke, M., and Lin, X. (2011) Rare-variant association testing for sequencing data with the sequence kernel association test. Am. J. Hum. Genet. 89, 82-93. DOI: http://dx.doi.org/10.1016/j.ajhg.2011.05.029{10.1016/j.ajhg.2011.05.029}.

See Also

GenotypeMatrix

Examples

## create a toy example
A <- matrix(rbinom(50, 2, prob=0.2), 5, 10)
pos <- sort(sample(1:10000, ncol(A)))

## compute some unweighted kernels
computeKernel(A, kernel="linear.podkat", pos=pos, width=100)
computeKernel(A, kernel="localsim.podkat", pos=pos, width=100)
computeKernel(A, kernel="linear.SKAT")

## compute some weighted kernels
MAF <- colSums(A) / (2 * nrow(A))
weights <- betaWeights(MAF)
computeKernel(A, kernel="linear.podkat", pos=pos, weights=weights)
computeKernel(A, kernel="linear.SKAT", weights=weights)
computeKernel(A, kernel="localsim.SKAT", weights=weights)