corrected_cs_bhat {corrcoverage}

Corrected credible set using estimated effect sizes and their standard errors

Description

Corrected credible set using estimated effect sizes and their standard errors

Usage

corrected_cs_bhat(
  bhat,
  V,
  N0,
  N1,
  Sigma,
  W = 0.2,
  lower = 0,
  upper = 1,
  desired.cov,
  acc = 0.005,
  max.iter = 20,
  pp0min = 0.001
)

Arguments

bhat

Estimated effect sizes

V

Prior variance of estimated effect sizes

N0

Number of controls

N1

Number of cases

Sigma

Correlation matrix of SNPs

W

Prior for the standard deviation of the effect size parameter, beta (default 0.2)

lower

Lower threshold (default = 0)

upper

Upper threshold (default = 1)

desired.cov

The desired coverage of the causal variant in the credible set

acc

Accuracy of corrected coverage to desired coverage (default = 0.005)

max.iter

Maximum iterations (default = 20)

pp0min

Only average over SNPs with pp0 > pp0min

Value

List of variants in credible set, required threshold, the corrected coverage and the size of the credible set

Author(s)

Anna Hutchinson

Examples

 # this is a long running example

# In this example, the function is used to find a corrected 95% credible set
# using bhats and their standard errors, that is the smallest set of variants
# required such that the resultant credible set has coverage close to (/within
# some accuracy of) the "desired coverage" (here set to 0.95). Max.iter parameter
# defines the maximum number of iterations to try in the root bisection algorithm,
# this should be increased to ensure convergence to the desired coverage, but is set
# to 1 here for speed (and thus the resultant credible set will not be accurate).

set.seed(18)
nsnps <- 100
N0 <- 500 # number of controls
N1 <- 500 # number of cases

# simulate fake haplotypes to obtain MAFs and LD matrix
## generate example LD matrix
library(mvtnorm)
nsamples = 1000

simx <- function(nsnps, nsamples, S, maf=0.1) {
    mu <- rep(0,nsnps)
    rawvars <- rmvnorm(n=nsamples, mean=mu, sigma=S)
    pvars <- pnorm(rawvars)
    x <- qbinom(1-pvars, 1, maf)
}

S <- (1 - (abs(outer(1:nsnps,1:nsnps,`-`))/nsnps))^4
X <- simx(nsnps,nsamples,S)
LD <- cor2(X)
maf <- colMeans(X)

varbeta <- Var.data.cc(f = maf, N = N0 + N1, s = N1/(N0+N1))

bhats = rnorm(nsnps,0,0.2) # log OR

names(bhats) <- seq(1,length(bhats))

corrected_cs_bhat(bhat = bhats, V = varbeta, N0, N1, Sigma = LD, desired.cov = 0.9, max.iter = 1)
# max.iter set low for speed, should be set to at
# least the default to ensure convergence to desired coverage