lrcde: lrcde

In ERGlass/lrcde.dev: Linear Regression Cell Type-Specific Differential Expression Power Analysis

Description

Cell type-specific differential expression detection given heterogeneous gene expression matrix, cell type-specific cell proportions, and a vector of group assignment (e.g., case-control study).

Usage

lrcde(het.sub, cell.props, groups, output.file = "LRCDE_power_analysis.csv",
  VERBOSE = TRUE, method = "dual", direction = "two.sided")

Arguments

het.sub	the samples (rows) by genes (columns) heterogeneous gene expression matrix. Should contain non-median-centered, non-standardized, positive values. log2-transformation recommended. Required
cell.props	the cell proportion matrix, cell types (rows) by samples (columns). The proportion should be relative, e.g., sum up to ~1 per sample. Required
groups	a vector of 1's and 2's indicating group assignment. Should correspond to the sample order in het.sub and cell.props. Required
output.file	file name to output the comma-separated results. Default: LRCDE_power.analysis.csv.
VERBOSE	boolean, whether to output progess to console. Default is TRUE.
method	cpecifies which type of regression deconvolution to perform. Only "dual" method (deconvolution of each group separately) is implemented in the current version (default). Future methods may include "ridge" and/or "single".
direction	the type of test to perform. Should be one of "two.sided" (default), "up", or "down".

Details

Each row of the results data.frame contains cell type-specific differential expression analysis statistics. Each column contains:

site

gene index, each gene is tested for differential expression in each cell type

base

cell type-specific gene expression estimate in _group 1_ (e.g., controls)

case

cell type-specific gene expression estimate in _group 2_ (e.g., cases)

diff.est

cell type-specific gene expression difference estimate. Larger differences are of interest, given sufficient power

mse.control/case

cell type-specific mean squared error gene expression estimatefor group 1/2, respectively

cell

cell type-specific index

cell.sd

cell type-specific standard deviation across samples

kappa.1/2

group-specific condition number for the cell proportion matrixes, for group 1/2, respectively

t.crit

t-statistics cutoff to reject the null hypothesis of no cell type-specific differential expression

t.stat/p.val.t

observed t-statistics/p-value, respectively, for the cell type-specific differential expression analysis. Not corrected for multiple testing. ToDo: use '1 - p' for negative diff.est

se.1/se.2

standard error of the cell type-specific gene expression estimates for group 1/2, respectively

t.power

power of the cell type-specific differential analysis. Non-siginficant power defaults to p.val.t, power at p.val.t < 0.05 indicates potentially significant cell type-specific differences. Larger power (> 0.8) corresponds to more significant differences.

Value

A list with three elements. The first contains a data.frame of analysis results. The second contains a list of parameters supplied to the lrcde function.

Author(s)

Edmund R Glass, [e-mail], Mikhail G Dozmorov, [e-mail]

References

https://github.com/ERGlass/lrcde.dev

Examples

## Not run: 
library("lrcde")
library(CellMix)
library(GEOquery)

gedDataInfo() # Data sets in CellMix to select from
mix <- ExpressionMix("GSE19830") # Example dataset with gene expression form mixture of tissues
p.sub <- subset(pData(mix), select = c(Liver, Brain, Lung, characteristics_ch1))

het <- t(exprs(mix))  # Heterogeneous expression matrix
cell.props <- t(coef(mix)) # Cell proportions matrix

apply(cell.props, 2, sd) # SD of cell proportions should not be 0
# Cell proportion standard deviation > 0.06 across samples should
# produce reasonable differential detection power.  
apply(cell.props, 1, sum)  # These sum perfectly to 1.

# Create (artificially) two groups, Brain and Lung vs. Liver
p.sub.1 <- p.sub[(p.sub$Brain + p.sub$Lung) > (p.sub$Liver), ]
# Heterogeneous matrixes for each group
het.1 <- het[rownames(het) %in% rownames(p.sub.1), ]
het.2 <- het[!(rownames(het) %in% rownames(p.sub.1)), ]
het2use <- rbind(het.1, het.2) # Combined matrix
# Cell proportions for each group
cells.1 <- cell.props[rownames(cell.props) %in% rownames(p.sub.1), ]
cells.2 <- cell.props[!(rownames(cell.props) %in% rownames(p.sub.1)), ]
cell.props <- rbind(cells.1, cells.2) # Combined cell proportions
# Group vector
groups <- c(rep(1, dim(het.1)[1]), rep(2, dim(het.2)[1]))

# Now apply the lrcde function in order to deconvolve cell
# type-specific expression and perform power analysis on differential
# expression detection:

power.analysis.df <- lrcde(het2use, 
                           cell.props, 
                           groups, 
                           output.file = "LRCDE_power_analysis_results.csv", 
                           VERBOSE = TRUE, 
                           method = "dual", 
                           direction = "two.sided")

## End(Not run)

ERGlass/lrcde.dev documentation built on May 6, 2019, 3:09 p.m.