EPSLASSO: Test for high-dimensional sparse regression for extreme...

In xu1912/EPSLASSO: Sparse Regression for data under extreme phenotype sampling

Test for high-dimensional sparse regression for extreme phenotype sampling data.

Description

This function tests the effects of predictors modelled jointly for high-dimensional extreme phenotype sampling data. The result returns 3 items: pvals, beta, and sigma. pvals contains score, wald, and partial likelihood ratio test p-value. beta is the point estimate of the regression coefficient. sigma is the estimate of random noise.

Usage

epsLasso(X, Y, c1, c2, lam0 = NULL, m_w = "lso", scal.x = TRUE,
  center.y = TRUE, paral = FALSE, paral_n = NULL, resol = 1.3,
  tol = 0.001, maxTry = 10, verbose = TRUE)

Arguments

X	Matrix of predictors. Required.
Y	Trait values. Required.
c1	Right censored point. Required.
c2	Left censored point. Required.
lam0	A sequence of lambda values. Default is the sequence used in GLMNET.
m_w	Methods used to estimate W matrix. Default is "lso" for LASSO solution using glmnet. Another method is "dzg" for Danzig-type estimator.
scal.x	Scale matrix X or not. Default is TRUE.
center.y	Center Y or not. Default is TRUE.
paral	Parallel computing indicator. Default is FALSE, not using parallel.
paral_n	Number of cores that are used for parallel computing. Default is NULL. When paral is TRUE, default is the number of system available cores - 1.
resol	The refining step when m_w="dzg". Default is 1.3. A large resol results in faster convergence speed, but rough solutions.
tol	The convergence threshold for refining when m_w="dzg". Default is 0.001.
maxTry	The maximum refining steps when m_w="dzg". Default is 10.
verbose	Print debugging info or not.

Examples

library(mvtnorm)
sd=1
n=100
p1=0.2
p2=0.2
p=100
nc=10
eff=0.5

beta_eff=c(rep(eff,nc),rep(0,p-nc))

cov_m=diag(p)
X_b=rmvnorm(n,mean=c(rep(1,p/2),rep(2,p/2)), sigma=cov_m)
Y_b=X_b%*%beta_eff+rnorm(n,0,sd)
sample_threshold_low=ceiling(n*p1)
sample_threshold_up=ceiling(n*p2)

ind_sample = order(Y_b)[c(1:sample_threshold_low,(n-sample_threshold_up+1):n)]
Y_p = Y_b[ind_sample]
c1 = min(Y_b[ind_sample[-(1:sample_threshold_low)]])
c2 = max(Y_b[ind_sample[1:sample_threshold_low]])
X_p = X_b[ind_sample,]

res=epsLasso(X_p,Y_p,c1,c2)
res

xu1912/EPSLASSO documentation built on May 10, 2022, 11:23 a.m.