mcpsum: Function to obtain MCP estimates of a regression problem...

In SeojinHwang/scadsum: nonconvex penalty with summary statistics and a reference panel

Function to obtain MCP estimates of a regression problem given summary statistics and a reference panel

Description

Function to obtain MCP estimates of a regression problem given summary statistics and a reference panel

Usage

mcpsum(
  cor,
  bfile,
  lambda = exp(seq(log(0.001), log(0.1), length.out = 20)),
  shrink = 0.9,
  gamma = 3,
  thr = 1e-04,
  init = NULL,
  trace = 0,
  maxiter = 10000,
  blocks = NULL,
  keep = NULL,
  remove = NULL,
  extract = NULL,
  exclude = NULL,
  chr = NULL,
  mem.limit = 4 * 10^9,
  chunks = NULL,
  cluster = NULL
)

Arguments

cor	A vector of correlations (r)
bfile	PLINK bfile (as character, without the .bed extension)
lambda	A vector of \lambdas (the tuning parameter)
shrink	The shrinkage parameter s for the correlation matrix R
gamma	The tuning parameter \gamma
thr	convergence threshold for \beta
init	Initial values for \beta as a vector of the same length as cor
trace	An integer controlling the amount of output generated.
maxiter	Maximum number of iterations
blocks	A vector to split the genome by blocks (coded as c(1,1,..., 2, 2, ..., etc.))
keep	samples to keep
remove	samples to remove
extract	SNPs to extract
exclude	SNPs to exclude
chr	a vector of chromosomes
mem.limit	Memory limit for genotype matrix loaded. Note that other overheads are not included.
chunks	Splitting the genome into chunks for computation. Either an integer indicating the number of chunks or a vector (length equal to cor) giving the exact split.
cluster	A cluster object from the parallel package for parallel computing

Details

A function to find the minimum of \beta in

f(\beta)=\beta'R\beta - 2\beta'r + 2\rho_{mcp}(\beta;\lambda,\gamma)

where

R=(1-s)X'X/n + sI

is a shrunken correlation matrix, with X being standardized reference panel. s should take values in (0,1]. r is a vector of correlations. keep, remove could take one of three formats: (1) A logical vector indicating which indivduals to keep/remove, (2) A data.frame with two columns giving the FID and IID of the indivdiuals to keep/remove (matching those in the .fam file), or (3) a character scalar giving the text file with the FID/IID. Likewise extract, exclude can also take one of the three formats, except with the role of the FID/IID data.frame replaced with a character vector of SNP ids (matching those in the .bim file).

Value

A list with the following

lambda	same as the lambda input
gamma	same as the gamma input
beta	A matrix of estimated coefficients
conv	A vector of convergence indicators. 1 means converged. 0 not converged.
pred	=\sqrt(1-s)X\beta
loss	=(1-s)\beta'X'X\beta/n - 2\beta'r
fbeta	=\beta'R\beta - 2\beta'r + 2\lambda||\beta||_1
sd	The standard deviation of the reference panel SNPs
shrink	same as input
nparams	Number of non-zero coefficients

Note

Missing genotypes are interpreted as having the homozygous A2 alleles in the PLINK files (same as the --fill-missing-a2 option in PLINK).

SeojinHwang/scadsum documentation built on June 30, 2023, 10:52 p.m.