marg2con: Obtain conditional estimates from marginal estimates and a...

In vincent10kd/polygenic: Efficient and User-friendly P+T Polygenic Risk Score Creation with Summary Statistics

Obtain conditional estimates from marginal estimates and a correlation matrix

Description

Returns conditional estimates based on marginal estimates (such as public GWAS summary statistics) and a correlation matrix. Allows post-hoc adjustments for covariates without access to the original data. Among other things, this is useful for fine-mapping and for decorrelating variants to prevent double-counting in polygenic risk scores.

Usage

marg2con(
  marginal_coefs,
  X = NULL,
  covX = NULL,
  corX = NULL,
  MAF = NULL,
  N = NULL,
  marginal_se = NULL,
  estimate_se = FALSE,
  y = NULL,
  varY = NULL,
  ridge = FALSE,
  lambda = 0,
  binary = FALSE,
  caseprop = 0.25
)

Arguments

marginal_coefs	A numeric vector of k marginal coefficient estimates from linear or logistic regression models.
X	An n*k matrix of predictors, with k equal to the length of marginal_coefs. Is not necessary when covX or corX and MAF are provided.
covX	A k*k covariance matrix.
corX	A k*k correlation matrix.
MAF	A numeric vector of k minor allele frequencies.
N	A single constant representing the sample size of the GWAS from which the marginal estimates were derived.
marginal_se	A numeric vector of k marginal standard error estimates from linear or logistic regression models.
estimate_se	If TRUE and if marginal_se is not NULL, returns conditional standard error estimates.
y	A continuous or binary response vector. If provided together with X, standard OLS is fit to obtain the estimates.
varY	The variance of the response vector. Is estimated if not provided.
ridge	If TRUE, applies a ridge penalty.
lambda	The parameter controlling the degree of (ridge) regularization.
binary	If TRUE, assumes that marginal_coefs are log odds ratios and that the standard errors should be estimated based on a different variance approximation for the response vector.
caseprop	The proportion of the response vector that represents 'cases', i.e. the proportion of the binary-valued vector that is equal to 1.

Value

A data.frame of conditional regression estimates, including standard errors and p-values depending on the input arguments.

Examples

marg_coefs <- rbind(summary(lm(Sepal.Length~Petal.Width, iris))$coef[2,1:2],
summary(lm(Sepal.Length~Petal.Length, iris))$coef[2,1:2],
summary(lm(Sepal.Length~Sepal.Width, iris))$coef[2,1:2])

covX <- cov(iris[c('Petal.Width','Petal.Length','Sepal.Width')])
N <- nrow(iris)
marg2con(marginal_coefs = marg_coefs[,1],
        covX = covX,
        N = N,
        marginal_se = marg_coefs[,2],
        estimate_se = TRUE)

vincent10kd/polygenic documentation built on Feb. 25, 2024, 10:17 a.m.