CC.Missing: Estimate latent covariates and coefficients of interest

In chrismckennan/MetabMiss: Estimate and account for non-ignorable missingness mechanisms and latent confounding.

Description

Estimate latent covariates and estimate/do inference on the coefficients of interest in a multivariate linear model with stabilized inverse probability weighting (sIPW) using estimated missingness mechansim.

Usage

CC.Missing(
  Y,
  X,
  Z = NULL,
  K = NULL,
  Miss.Mech,
  ind.samples = NULL,
  max.miss.perp = 0.5,
  max.miss.image = 0.5,
  BH.min = NULL,
  method = c("sIPW", "IPW"),
  include.updates = T,
  est.Beta = T,
  return.nuis = F,
  refine.C = F,
  p.refine.both = F,
  return.all = F,
  return.mu = F
)

Arguments

Y	a p x n data matrix of log2-transformed metabolite intensities, where p = #of metabolites and n = #of samples. Missing values should be left as NA.
X	a n x d matrix of covariates of interest (i.e. disease status).
Z	a n x r matrix of observed nuisance covariates (i.e. the intercept, observed technical factors, etc.)
K	The number of latent covariates (i.e. C is a n x K matrix). If unspecified, it is estimated using sva::num.sv applied to the metabolites with complete data.
Miss.Mech	The missingness mechansim object returned by EstimateMissing.
ind.samples	A logical or numeric vector of samples to be considered in the analysis. For example, if disease status were only measured in a subset of the patients, this would be the samples with a recorded disease status. Default is no missing samples.
max.miss.perp	Maximum fraction of missing data a metabolite is allowed to have to be used to calculate the part of C perpendicular to X. Defaults to 0.5.
max.miss.image	Maximum fraction of missing data a metabolite is allowed to have to be used to calculate the part of C in the image of X. Defaults to 0.5.
est.Beta	A logical indicating whether or not to estimate/do inference on coefficients of interest. If F, only the latent covariates are estimated. Default, and recommended value, is T.

Value

A list

C.iter	The estimate of the n x K matrix of latent covariates.
Beta.iter	The estimate of the p x d matrix of coefficients of interest.
p.t.iter	The p x d matrix of p-values for the coefficients of interest.
Var.beta.iter	A length p list of the d x d estimates for Var(Beta.iter)
t.iter	A p x d matrix of t-statistics, defined as Beta.iter/SE(Beta.iter)
p.f.iter	A length p vector of F-statistic p-values for the null hypothesis that X has no effect on metabolite intensity. It is only returned if d > 1.
L	The estimate of the p x K matrix of coefficients for the latent covariates.
Beta.naive	The estimate of the p x d matrix of coefficients of interest that ignores C. This should ONLY be used for comparison.
p.t.naive	The p x d matrix of p-values for the coefficients of interest that ignore C. This should ONLY be used for comparison.

chrismckennan/MetabMiss documentation built on March 1, 2020, 10:03 p.m.