Support functions for MethylCIBERSORT

Custom scripts used within "Pediatric Pan-CNS Tumor Analysis of Immune-cell Infiltration Identifies Correlates of Antitumor Immunity" by Grabovska et al. (2020)

• Overview
• Documentation
• feature.select.new
• make.synth.mix
• System Requirements
• Hardware requirements
• Software Requirements
• Usage Guide
• Issues

Overview

The functions provided serve two main purposes:

• feature.select.new(): facilitate and extend the 'feature selection' functionality of the MethylCIBERSORT R software package
• make.synth.mix(): facilitate benchmarking of resulting signature matrix by generating synthetic mixtures of test populations with known input proportions

Documentation

feature.select.new

This function mimics the functionality of FeatureSelect.V4() from MethylCIBERSORT 0.2.1 written by Dr Ankur Chakravarthy

Default function call:

feature.select.new(MaxDMPs = 100, deltaBeta = 0.2, useM = FALSE, CellLines.matrix = NULL, export = TRUE, export.fit = TRUE, export.cpg = TRUE, sigName = "methylCibersort", Stroma.matrix = NULL, Phenotype.stroma = NULL, FDR = 0.01, silent = TRUE)

Parameters:

MaxDMPs: maximum differentially methylated probes to use, takes n/2 from top & n/2 from bottom

deltaBeta: cutoff for the minimum difference between pairwise groups by delta-beta

useM: specify whether conversion to M-values should be done before carrying out feature selection

CellLines.matrix: input matrix for cell line 'cancer' data

export: save a table of signature results

export.fit: whether to export the limma fit object during the feature selection

export.cpg: whether to export a table of the CpG probes selected alongside the name of the population which they were selected against

sigName: name appended to start of filename

Stroma.matrix: matrix of betas for populations

Phenotype.stroma: pheno that corresponds to Stroma.matrix

FDR: FDR cutoff

silent: run function without returning output

Typical outputs:

• .txt file of the resulting signature matrix
• .rds file of the limma fit
• .txt file of CpGs selected and the comparison from which they were chosen
• .txt file of the Illumina/minfi CpG annotations for the relevant CpGs

Note on functionality

The typical function workflow is shown below, various set and optional arguments are shown in bold:

make.synth.mix

This function creates a matrix of proportions for a given set of data populations and generates weighted means based on those proportions for a given matrix of values, typically methylation array beta-values.

Default function call:

make.synth.mix(input.data = NULL, pop.factor = NULL, pop.rows = 100, output.dir = getwd(), output.name = gsub("-", "_", Sys.Date()), n.cores = 1)

Parameters:

input.data: a matrix of data, typically beta-values; rows = features (probes), columns = observations (samples)

pop.factor: a factor with levels describing the populations in input.data columns

pop.rows: numeric specifying how many rows of proportions to generate for each population

output.dir: location to save the resulting files

output.name: name to append to the resulting filenames

n.cores: set to >1 to run on multiple cores in parallel using parallel::mcapply()

Typical outputs:

• .txt file of the resulting proportions matrix
• .txt file of the resulting mixtures

Note on functionality:

The function populates the proportion table by column, with each population being assigned a set sequence of probabilities. A typical example for 3 populations, 6 rows per population:

System Requirements

Hardware Requirements

Functions provided here are compatible with any standard computer with enough RAM to support the in-memory operations. The functions are supported on any operating system which supports MethylCIBERSORT 0.2.1

Software Requirements

Functions provided import and depend on a number of R packages. Functionality has been tested on R 3.5.3 with Ubuntu 16.04.5 LTS

R Dependencies

MethylCIBERSORT 0.2.1 was obtained by correspondence from Dr Ankur Chakravarthy.

magrittr
matrixStats
BiocGenerics
limma
dplyr
minfi
parallel
minfiData
MethylCIBERSORT
IlluminaHumanMethylation450kmanifest

Usage Guide:

Example Outputs

Example outputs generated from running the below usage examples included for comparison within Example Outputs

The final signature generated as part of our study is available to download - OPT2_0.2_200_SigEdit.txt

Feature selection example:

source("./feature_select_WORKING_2019.R")
require(FlowSorted.Blood.450k)
require(IlluminaHumanMethylation450kmanifest)
idx <- which(FlowSorted.Blood.450k$CellType %in% c("Bcell","CD4T","CD8T","NK"))
bvals <- getBeta(FlowSorted.Blood.450k[, idx])
getwd() ## outs stored in working directory
a <- Sys.time()
feature.select.new(Stroma.matrix = bvals,
                   Phenotype.stroma = as.factor(FlowSorted.Blood.450k$CellType[idx]),
                   sigName = "test")
b <- Sys.time()
b-a
# Time difference of 1.55212 mins

Generating synthetic mixtures example:

source("./synth_mix_2019.R")
require(FlowSorted.Blood.450k)
idx <- which(FlowSorted.Blood.450k$CellType %in% c("Bcell","CD4T","CD8T","NK"))
bvals <- getBeta(FlowSorted.Blood.450k[, idx])
b.sig <- read.delim("./test_0.2_100_Signature.txt", row.names = 1, header = TRUE)
bvals <- bvals[rownames(b.sig), ]
all(rownames(bvals)==rownames(b.sig))
a <- Sys.time()
make.synth.mix(input.data = bvals,
               pop.factor = as.factor(FlowSorted.Blood.450k$CellType[idx]),
               pop.rows = 10,
               n.cores = 1,
               output.name = "test")
b <- Sys.time()
b-a
## Time difference of 1.76159 secs

Example CIBERSORT analysis

a <- Sys.time()
source("./CIBERSORT.R") ## code available under license upon request from https://cibersort.stanford.edu/
results <- CIBERSORT(sig_matrix = "./test_0.2_100_Signature.txt",
                     mixture_file = "./test_synth_mix.txt",
                     perm = 1000,
                     QN = FALSE,
                     absolute = FALSE,
                     abs_method = 'sig.score')
b <- Sys.time()
b-a
## Time difference of 10.40594 mins

## the above is not a real world example however we provide here the full signature generated and used in our study and we typically run something like this...
# results <-CIBERSORT(sig_matrix = "./OPT2_0.2_200_SigEdit.txt", ## the signature from our study
#                    mixture_file = "a_mixture_file_from_CNS_tumours.txt", ## i.e. make your own beta matrix
#                    perm = 1000,
#                    QN = F,
#                    absolute = F,
#                    abs_method = 'sig.score')