FastMix: A wrapper for FastMix pipeline.

In terrysun0302/FastMix: An OLS-based approach for fitting LMER with applications to gene expression deconvolution analysis

Description

A wrapper for the main function ols.eblup.trim() to conduct deconvolution gene expression analysis with matching cell proportions.

Usage

FastMix(GeneExp, CellProp, Demo, random="all", include.demo=TRUE, cov_matrix = NULL, ...)

Arguments

GeneExp	'GeneExp' is a m by n dimensional gene expression matrix, where m is the number of genes, and n is the number of subjects.
CellProp	'CellProp' is a n by K dimensional matrix of cell proportions, where K is the number of cell proportions.
random	random is an index vector that specifies which variable(s) requires random effects – by default, random="all", which means that all covariates are paired with a random effect.
Demo	Demo is a n by p dimensional matrix of clinical and demographic variables to be tested, where p is the number of covariates.
include.demo	Whether the demographical covariates should be included as the main effects in the model or not. Default to TRUE.
cov_matrix	The sample-specific weight.
...	Additional parameters passed to ols.eblup.trim(). It includes the following useful options independent: independent specifies the correlation structure among random effects. The default value is TRUE, which means that all random effects are assumed to be independent. trim: trim is the trimming percentage when accounting for outliers. Default valie is 0.5 (50%). test: the test method for DEGs. "1" is Gaussian mixture model, "2" is Anderson-darling normal test. Default value is "1". robust: robust specifies whether robust covariance estimation is implemented and which method to use: "FALSE" for non-robust estimation; "mcd" for the MCD algorithm of Rousseeuw and Van Driessen; "weighted" for the Reweighted MCD; "donostah" for the Donoho-Stahel projection based estimator; "pairwiseQC" for the orthogonalized quadrant correlation pairwise estimator. All these algorithms come from the R package 'robust'. The default value is robust="FastMix", which is the proposed trimming method. trim.fix: Whether only consider trimmed subjects in fix effect estiamtion. The default value is FALSE. weight_matrix: The pre-specified weight structure among subjects.

Value

fixed.results	the estimated fix effects and their p-values. They are overall effects shared by all genes.
beta.mat	estimated linear coefficients for individual genes.
GeneExp.fitted	fitted gene expressions.
sigma.beta	the estimated covariance matrix of the fixed effects.
VC	variance component estimation. The first column is the one for common random error. The second column is the one for random effects.
cov	???
var.epsilon	the variance of the i.i.d. noise.
var.eblup.mean	the average of the variance of gamma.hat based on the EBLUP estimator. Note that in general, each gamma.hat.i has its own covariance matrix; so var.eblup.mean is provided only as a rough debugging tool.
eta	the chi-square type statsitics used for p-value calculation.
re.pvalue	the overall p-value for detecting outliers in random effects.
re.ind.pvalue	the individual p-value for outlier detection for each random effect.
out_idx	the potential covariates with outliers when robust = "FastMix. It is NULL when robust != "FastMix"

Author(s)

Hao Sun

Examples

## load the data example
data(dat_train)

## fit the model by default parameters
mod1 <- FastMix(dat_train$GeneExp, dat_train$CellProp, dat_train$Demo)

## some variants: only assign random effects to a subset of covariates, and uses non-robust method to estimate the covariance structure
mod2 <- FastMix(dat_train$GeneExp, dat_train$CellProp, dat_train$Demo, random=c(1,2,10))

terrysun0302/FastMix documentation built on Nov. 14, 2019, 4:54 a.m.