or relatedness

computeVSIF

R Documentation

Computes variant-specific inflation factors

Description

Computes variant-specific inflation factors resulting from differences in variances and allele frequencies across groups pooled together in analysis.

Usage

computeVSIF(freq, n, sigma.sq)
computeVSIFNullModel(null.model, freq, group.var.vec)

Arguments

`freq`	A named vector or a matrix of effect allele frequencies across groups. Vector/column names are group names; rows (for a matrix) are variants.
`n`	A named vector of group sample sizes.
`sigma.sq`	A named vector of residual variances across groups.
`null.model`	A null model constructed with `fitNullModel`.
`group.var.vec`	A named vector of group membership. Names are sample.ids, values are group names.

Details

computeVSIF computes the expected inflation/deflation for each specific variant caused by differences in allele frequencies in combination with differences in residual variances across groups that are aggregated together (e.g. individuals with different genetic ancestry patterns). The inflation/deflation is especially expected if a homogeneous variance model is used.

computeVSIFNullModel uses the null model and vector of group membership to extract sample sizes and residual variances for each group. It then calls function computeVSIF to compute the inflation factors. The null model should be fit under a homogeneous variance model.

Value

`SE_true`	Large sample test statistic variances accounting for differences in residual variances.
`SE_naive`	Large sample test statistic variances (wrongly) assuming that all residual variances are the same across groups.
`Inflation_factor`	Variant-specific inflation factors. Values higher than 1 suggest inflation (too significant p-value), values lower than 1 suggest deflation (too high p-value).

Author(s)

Tamar Sofer, Kenneth Rice

References

Sofer, T., Zheng, X., Laurie, C. A., Gogarten, S. M., Brody, J. A., Conomos, M. P., ... & Rice, K. M. (2020). Population Stratification at the Phenotypic Variance level and Implication for the Analysis of Whole Genome Sequencing Data from Multiple Studies. BioRxiv.

Examples

  n <- c(2000, 5000, 100)
  sigma.sq <- c(1, 1, 2)
  freq.vec <- c(0.1, 0.2, 0.5)
  names(freq.vec) <- names(n) <- names(sigma.sq) <- c("g1", "g2", "g3") 
  res <- computeVSIF(freq = freq.vec, n, sigma.sq)
  
  freq.mat <- matrix(c(0.1, 0.2, 0.5, 0.1, 0.01, 0.5), nrow = 2, byrow = TRUE)
  colnames(freq.mat) <- names(sigma.sq)
  res <- computeVSIF(freq = freq.mat, n, sigma.sq)


  library(GWASTools)
  n <- 1000
  set.seed(22)
  outcome <- c(rnorm(n*0.28, sd =1), rnorm(n*0.7, sd = 1), rnorm(n*0.02, sd = sqrt(2)) )
  dat <- data.frame(sample.id=paste0("ID_", 1:n),
                    outcome = outcome,
                    b=c(rep("g1", n*0.28), rep("g2", n*0.7), rep("g3", n*0.02)),
                    stringsAsFactors=FALSE)
  dat <- AnnotatedDataFrame(dat)
  nm <- fitNullModel(dat, outcome="outcome", covars="b", verbose=FALSE)
  freq.vec <- c(0.1, 0.2, 0.5)
  names(freq.vec) <- c("g1", "g2", "g3") 
  group.var.vec <- dat$b
  names(group.var.vec) <- dat$sample.id
  
  res <- computeVSIFNullModel(nm, freq.vec, group.var.vec)
  
  freq.mat <- matrix(c(0.1, 0.2, 0.5, 0.1, 0.01, 0.5), nrow = 2, byrow = TRUE)
  colnames(freq.mat) <- c("g1", "g2", "g3")
  res <- computeVSIFNullModel(nm, freq.mat, group.var.vec)

UW-GAC/GENESIS documentation built on May 16, 2024, 1:10 p.m.