tcgsaseq (version 2.0.5)

vc_score_h: Computes variance component test statistic for homogeneous trajectory

Description

This function computes an approximation of the variance component test for homogeneous trajectory based on the asymptotic distribution of a mixture of \(\chi^{2}\)s using Davies method from davies

Usage

vc_score_h(y, x, indiv, phi, w, Sigma_xi = diag(ncol(phi)), na_rm = FALSE)

Arguments

y

a numeric matrix of dim g x n containing the raw or normalized RNA-seq counts for g genes from n samples.

x

a numeric design matrix of dim n x p containing the p covariates to be adjusted for

indiv

a vector of length n containing the information for attributing each sample to one of the studied individuals. Coerced to be a factor.

phi

a numeric design matrix of size n x K containing the K longitudinal variables to be tested (typically a vector of time points or functions of time)

w

a vector of length n containing the weights for the n samples, corresponding to the inverse of the diagonal of the estimated covariance matrix of y.

Sigma_xi

a matrix of size K x K containing the covariance matrix of the K random effects corresponding to phi.

na_rm

logical: should missing values (including NA and NaN) be omitted from the calculations? Default is FALSE.

Value

A list with the following elements:

• score: approximation of the set observed score

• q: observation-level contributions to the score

• q_ext: pseudo-observations used to compute covariance taking into account the contributions of OLS estimates

• gene_scores: approximation of the individual gene scores

See Also

davies

Examples

# NOT RUN {
rm(list=ls())
set.seed(123)

##generate some fake data
########################
ng <- 100
nindiv <- 30
nt <- 5
nsample <- nindiv*nt
tim <- matrix(rep(1:nt), nindiv, ncol=1, nrow=nsample)
tim <- cbind(tim, tim^2)
sigma <- 5
b0 <- 10

#under the null:
beta1 <- rnorm(n=ng, 0, sd=0)
#under the (heterogen) alternative:
beta1 <- rnorm(n=ng, 0, sd=0.1)
#under the (homogen) alternative:
beta1 <- rnorm(n=ng, 0.06, sd=0)

y.tilde <- b0 + rnorm(ng, sd = sigma)
y <- t(matrix(rep(y.tilde, nsample), ncol=ng, nrow=nsample, byrow=TRUE) +
      matrix(rep(beta1, each=nsample), ncol=ng, nrow=nsample, byrow=FALSE)*matrix(rep(tim, ng),
                                                            ncol=ng, nrow=nsample, byrow=FALSE) +
      matrix(rnorm(ng*nsample, sd = sigma), ncol=ng, nrow=nsample, byrow=FALSE)
      )
myindiv <- rep(1:nindiv, each=nt)
x <- cbind(1, myindiv/2==floor(myindiv/2))
myw <- matrix(rnorm(nsample*ng, sd=0.1), ncol=nsample, nrow=ng)

#run test
score_homogen <- vc_score_h(y, x, phi=tim, indiv=myindiv,
                           w=myw, Sigma_xi=cov(tim))
score_homogen$score

score_heterogen <- vc_score(y, x, phi=tim, indiv=myindiv,
                           w=myw, Sigma_xi=cov(tim))
score_heterogen$score

scoreTest_homogen <- vc_test_asym(y, x, phi=tim, indiv=rep(1:nindiv, each=nt),
                                 w=matrix(1, ncol=ncol(y), nrow=nrow(y)), Sigma_xi=cov(tim),
                                 homogen_traj = TRUE)
scoreTest_homogen$set_pval
scoreTest_heterogen <- vc_test_asym(y, x, phi=tim, indiv=rep(1:nindiv, each=nt),
                                   w=matrix(1, ncol=ncol(y), nrow=nrow(y)), Sigma_xi=cov(tim),
                                   homogen_traj = FALSE)
scoreTest_heterogen$set_pval

# }