cor_shared: cor_shared

In jamesliley/cFDR-common-controls: Estimation of the Bayesian conditional false discovery rate (cFDR) allowing shared controls

Description

Compute correlation between z values arising from shared controls

Usage

cor_shared(n0i, n0j, overlap, n1i, n1j)

Arguments

n0i	Number of controls in GWAS for principal phenotype
n0j	Number of controls in GWAS for conditional phenotype
overlap	Number of controls shared between studies
n1i	Number of cases in GWAS for principal phenotype
n1j	Number of cases in GWAS for conditional phenotype

Details

Uses a formula from Zaykin et al (2010)

Value

Asymptotic correlation between z scores.

Author(s)

James Liley

Examples

n0i = 1500; n0j = 1200; n1i = 800; n1j = 700;
overlap = 1000;
cor_shared(n0i,n0j,n1i,n1j,overlap)


# Simulation
n_snps = 10000; sim_maf = runif(n_snps,min=0.1,max=0.5) # simulated MAF
sim_con_i = matrix(runif(n_snps*n0i),n_snps,n0i)<sim_maf # simulated genotypes for controls for phenotype i
sim_con_j = matrix(runif(n_snps*n0j),n_snps,n0j)<sim_maf; sim_con_j[,1:overlap]=sim_con_i[,1:overlap] # simulated genotypes for controls for phenotype j, with shared controls
sim_case_i = matrix(runif(n_snps*n1i),n_snps,n1i)<sim_maf; sim_case_j = matrix(runif(n_snps*n1j),n_snps,n1j)<sim_maf # simulated genotypes for cases
om0i=rowMeans(sim_con_i); om0j=rowMeans(sim_con_j); om1i = rowMeans(sim_case_i); om1j = rowMeans(sim_case_j) # observed minor allele frequencies
or_i = om0i*(1-om1i)/(om1i*(1-om0i)); or_j = om0j*(1-om1j)/(om1j*(1-om0j)) # Odds ratios
se_i = sqrt( (1/(2*om0i*n0i)) + (1/(2*(1-om0i)*n0i)) + (1/(2*om1i*n1i)) + (1/(2*(1-om1i)*n1i)) )
se_j = sqrt( (1/(2*om0j*n0j)) + (1/(2*(1-om0j)*n0j)) + (1/(2*om1j*n1j)) + (1/(2*(1-om1j)*n1j)) ) # Standard errors
effect_i = log(or_i)/se_i; effect_j = log(or_j)/se_j # Effect sizes

cor(effect_i,effect_j)
cor_shared(n0i,n0j,n1i,n1j,overlap)

cor(abs(effect_i),abs(effect_j))
cor_shared(n0i,n0j,n1i,n1j,overlap)^2

jamesliley/cFDR-common-controls documentation built on May 18, 2019, 11:21 a.m.