SKAT_Null_Model_CEP: Fit the Null model for CEP-SKAT

In lin-lab/CEPSKAT: Finding rare variant associations for extreme phenotype samples.

Description

This function should be used in conjunction with the SKAT function of the SKAT package. A null model is fit and prepared to be passed to the SKAT function in order to perform the score test for association.

Usage

SKAT_Null_Model_CEP(formula, c1, c2, delta = 0.001, gamma0 = "default", MAXITERNUM = 20, data = NULL, n.Resampling = 0)

Arguments

formula	The null model to be fit.
c1	This is the smallest (lower bound) for the high extreme phenotypes. In other words c1 is the lower bound for the phenotypes sampled from the upper tail of the distribution.
c2	This is the largest (upper bound) for the low extreme phenotypes. In other words c2 is the upper bound for the phenotypes sampled from the lower tail of the distribution.
delta	Specifies speed and accuracy of convergence. Large values of delta converge faster but are less accurate, small values of delta converge slower but are more accurate.
gamma0	The initial guess for the MLEs. Default uses estimates from OLS.
MAXITERNUM	If the algorithm has not converged after this many iterations, the function exits. This protects against divergence of the Newton-Raphson algorithm.
data	Original data.frame with phenotype and covariates. Optional: only use if formula contains references to items in data.
n.Resampling	Defaults to 0. If resampling is to be used, this specifies how many times to resample.

Details

SKAT tests for an association between a SNP-set and phenotype through a score test where the model is fit under the null hypothesis of no association. This function fits this null model that is independent of genotype data. After fitting this model, the SKAT package can handle the output combined with the genotypes to produce p-values.

Value

An item of class "SKAT_Null_Model" is returned which can be interpreted by the SKAT function in the SKAT package.

Author(s)

Ian Barnett

References

Wu, M. C.*, Lee, S.*, Cai, T., Li, Y., Boehnke, M., and Lin, X. (2011) Rare Variant Association Testing for Sequencing Data Using the Sequence Kernel Association Test (SKAT). American Journal of Human Genetics, 89, 82-93. \ * contributed equally.

Examples

# Example with no association
G = cbind(rbinom(1000,2,.2),rbinom(1000,2,.15))
Y = rnorm(1000) # No association
ind_sample = order(Y)[c(1:250,751:1000)]
Y_extreme = Y[ind_sample]
G_extreme = G[ind_sample,]
c1 = min(Y[ind_sample[-(1:250)]])
c2 = max(Y[ind_sample[1:250]])
obj=SKAT_Null_Model_CEP(Y_extreme~1,c1,c2)
SKAT(G_extreme,obj)$p.value

# Example with high association
Y = G
ind_sample = order(Y)[c(1:250,751:1000)]
Y_extreme = Y[ind_sample]
G_extreme = G[ind_sample,]
c1 = min(Y[ind_sample[-(1:250)]])
c2 = max(Y[ind_sample[1:250]])
obj=SKAT_Null_Model_CEP(Y_extreme~1,c1,c2)
SKAT(G_extreme,obj)$p.value



## The function is currently defined as
function (formula, c1, c2, delta = 0.001, gamma0 = "default", 
    MAXITERNUM = 20, data = NULL, n.Resampling = 0) 
{
    out <- SKAT_TruncatedNormalCEPSolve(formula, c1, c2, delta, 
        gamma0, MAXITERNUM, data)
    res = out$Y - out$Mu
    pi_1 = out$W
    X1 = out$X1
    n <- length(res)
    id_include = 1:n
    res.out = NULL
    type.Resampling = NULL
    if (n.Resampling > 0) {
        res.out <- SKAT_Get_TruncatedNormal(length(res), out$Xa, 
            out$sigma, c1, c2, nSet = n.Resampling) - out$Mu
        res.out <- t(t(res.out) - colMeans(res.out))
    }
    re <- list(res = res, X1 = X1, res.out = res.out, out_type = "D", 
        n.Resampling = n.Resampling, type.Resampling = NULL, 
        id_include = id_include, mu = out$Mu, pi_1 = pi_1, coef = out$coef, 
        Xa = out$Xa)
    class(re) <- "SKAT_NULL_Model"
    return(re)
  }

lin-lab/CEPSKAT documentation built on May 29, 2019, 3:41 a.m.