Getting Started with deconvMe

Getting Started with deconvMe

Welcome to deconvMe! This vignette will help you get up and running with cell-type deconvolution for DNA methylation data using the package’s unified interface and included methods.

Installation

You can install deconvMe from GitHub using the pak package manager:

# install the `pak` package manager
install.packages("pak")

pak::pkg_install("omnideconv/deconvMe")

Example Usage

deconvMe can be applied directly to a methylSet from the minfi package, or you can apply each method separately on a beta matrix with Illumina CpG IDs.

Below, we demonstrate how to use the EpiDISH method with example data from minfi:

library(deconvMe)
library(minfi)
library(minfiData)

# use example data from Minfi
methyl_set <- minfiData::MsetEx
ratio_set <- minfi::ratioConvert(methyl_set)
beta_matrix <- minfi::getBeta(ratio_set)

# run EpiDISH for deconvolution of example data
result <- deconvMe::deconvolute(methyl_set = methyl_set, method = 'epidish')

Viewing the Results

The result of the deconvolution is a table with the estimated cell-type fractions for each sample. You can view it directly as a nicely formatted table:

knitr::kable(head(result), caption = "Estimated cell-type fractions for the first few samples")

Estimated cell-type fractions for the first few samples

	B	CD4T	CD8T	Eosino	Mono	Neutro	NK
5723646052_R02C02	0.1868416	0.1993816	0	0.1853980	0.3058855	0	0.1224934
5723646052_R04C01	0.1841371	0.1411284	0	0.2169230	0.3220845	0	0.1357269
5723646052_R05C02	0.0000000	0.2990338	0	0.1252629	0.3593454	0	0.2163580
5723646053_R04C02	0.0610074	0.1500589	0	0.2699428	0.4363476	0	0.0826434
5723646053_R05C02	0.1221793	0.1659357	0	0.2266461	0.3354384	0	0.1498004
5723646053_R06C02	0.0000000	0.1364639	0	0.2388252	0.4611588	0	0.1635522

For more details, see the package documentation and other vignettes!

Overview of Included Methods

Below is a template table summarizing the deconvolution methods included in deconvMe. This table lists each method, the type of algorithm it uses, the intended tissue types, and the type of data used to build the internal deconvolution reference.

Name	Algorithm Type(s)	Intended Tissue Type(s)	Reference Source Data
EpiDISH	Robust Partial Correlations (RPC), Constrained Projection (CP), Support Vector Regression (SVR)	Blood, Epithelial, Breast	purified blood, epithelial, and non-epithelial data, extended with DNAse Hypersensitivity Sites from Roadmap and ENCODE
Houseman	Constrained projection/Quadratic Programming (CP/QP)	Blood	Purified blood cell methylation profiles (IDOL-optimized)
MethylCC	Constrained Linear model based on DMRs	Blood (in deconvMe, can be extended to other tissues)	purified blood
MethylResolver	Least Trimmed Squares (LTS) regression	Blood, cancer	Purified blood methylation profiles (IDOL-optimized), extended with additional leukocytes
MethAtlas	non-negative least squares (NNLS)	Blood, immune, tissue-wide	Comprehensive atlas of purified tissue and immune cell methylomes

This table can be expanded as new methods are added to the package.