geneJAM: Cluster multiple traits via polygenic scores

In abuchardt/EdGwas: Joint regression Analysis of Multiple traits based on GENEtic relationships

Cluster multiple traits via polygenic scores

Description

This function clusters traits that share some genetic component via polygenic scores (PSs). It fits a sparse precision matrix via graphical lasso. The regularisation path is computed for the lasso penalty at a grid of values for the regularisation parameter rho. The clusters are used to specify the structure of the error covariance matrix of a GLS model and the feasible GLS estimator is used for estimating the unknown parameters in a linear regression model with a certain unknown degree of correlation between the residuals.

Usage

geneJAM(
  x,
  y,
  rho = NULL,
  nrho = ifelse(is.null(rho), 20, length(rho)),
  logrho = FALSE,
  rho.min.ratio = 0.001
)

Arguments

x	Input matrix, of dimension nobs x nvars of PGSs with nvars = nouts. Can be in sparse matrix format.
y	Quantitative response matrix, of dimension nobs x nouts.
rho	(Non-negative) regularisation parameter for lasso passed to glasso. rho=0 means no regularisation. Can be a scalar (usual) or a symmetric nouts by nouts matrix, or a vector of length nouts. In the latter case, the penalty matrix has jkth element sqrt(rho[j]*rho[k]).
nrho	The number of rho values - default is 40.
logrho	Logical flag for log transformation of the rho sequence. Default is logrho = FALSE.
rho.min.ratio	Smallest value for rho, as a fraction of rho.max, the (data derived) entry value (i.e. the smallest value for which all coefficients are zero) - default is 10e-04.

Details

...

Value

An object of class "geneJAM" is returned.

call	The call that produced this object.
xi	A matrix of intercepts of dimension nouts x length(rho)
beta	A matrix of coefficients for the PSs of dimension nouts x length(rho)
A	A length(rho) list of estimated adjacency matrices A of 0s and 1s, where A_ij is equal to 1 iff edges i and j are adjacent and A_ii is 0.
P	A length(rho) list of estimated precision matrices (matrix inverse of correlation matrices).
Sigma	A length(rho) list of estimated correlation matrices.
rho	The actual sequence of rho values used.
PS	Polygenic scores used. If scores = FALSE they are computed by ps.geneJAM
logrho	Logical flag for log transformation of the rho sequence. Default is logrho = FALSE.
nobs	Number of observations.
xiStderr	Standard error of coefficients xi.
betaStderr	Standard error of coefficients beta.
betaSD	Standard error of the mean of coefficients beta for clustered traits.
betaSD0	Standard error of the mean of all coefficients beta.
rho.min	Value of rho that gives minimum non-zero betaSD.

Examples

N <- 1000 #
q <- 10 #
p <- 1000 #
set.seed(1)
# Sample 1
X0 <- matrix(rbinom(n = N*p, size = 2, prob = 0.3), nrow=N, ncol=p)
B <- matrix(0, nrow = p, ncol = q)
B[1, 1:2] <- 1
B[3, 3] <- 2
y0 <- X0 %*% B + matrix(rnorm(N*q), nrow = N, ncol = q)
#y0 <- apply(y0, 2, scale)
beta <- ps.geneJAM(X0, y0)$beta
# Sample 2
X <- matrix(rbinom(n = N*p, size = 2, prob = 0.3), nrow=N, ncol=p)
y <- X %*% B + matrix(rnorm(N*q), nrow = N, ncol = q)
#y <- apply(y, 2, scale)
#Sigma <- diag(1, q)
#Sigma[1, 2] <- Sigma[2, 1] <- .8
#y <- X %*% B + MASS::mvrnorm(n = N, mu = rep(0, q), Sigma = Sigma)
x <- X %*% beta
#x <- MASS::mvrnorm(n = N, mu = rep(0, q), Sigma = diag(1, q))
###
pc <- geneJAM(x, y)

abuchardt/EdGwas documentation built on Nov. 28, 2022, 11:49 a.m.