In elmerfer/MIXTURE: MIXTURE: a nu-SVR based recursive feature extraction mdoel with noise constraints for molecular signature deconvolution of bulk tumor tissues

A noise constrained Recursive Feature Extraction algorithm for robust deconvolution of cell-types mixture from molecular signatures

Since the significant impact of immunotherapy in cancer, the estimation of the immune cell-type proportions present in a tumor becomes crucial. Currently, the deconvolution of the cell mixture content of a tumor is carried out by different analytic tools, yet the accuracy of inferred cell type proportions has room for improvement. We improve tumor immune environment characterization developing MIXTURE, an analytical method based on a noise constrained recursive variable selection for a support vector regression

How to install MIXTURE

install.packages("devtools")
library(devtools)
install_github("elmerfer/MIXTURE")

Running MIXTURE

In this example we will use the LM22 signature matrix as a 22 subject expression matrix cohort

library(MIXTURE)
##Load signature matrix
data(LM22)
##  Run the self test on LM22 signature
mix.test <- MIXTURE(expressionMatrix = LM22,          #N x ncol(signatureMatrix) gene expression 
                                                      ##matrix to evaluate 
                                                      ##rownames(M) should be the GeneSymbols
              signatureMatrix = LM22,                 #the gene signature matrix (W) such that 
                                                      ##M = W*betas' 
                                                      #(i.e the LM22 from Newman et al)
              functionMixture = nu.svm.robust.RFE,    #cibersort, nu.svm.robust.rfe, ls.rfe.abbas, 
              useCores = 10L,                         #cores for parallel processing
              verbose = TRUE,                         #TRUE or FALSE messages  
              nullDist = "PopulationBased" 
              ) 
# Showing the predicted proportions
head(GetMixture(mix.test)[,1:3])
# Showing the predicted absolute coefficients
head(GetMixture(mix.test, type = "absolute")[,1:3])
# Showing the slots names of the MIXTURE object
names(mix.test)

How to performe a molecular signature SelfTest analysis with MIXTURE ?

A self test analysis is intended to evaluate if the signature profiles can be accuratly predicted by using MIXTURE. This means that each signature profiles (Columns) will be assumed as a pure cell mixture profile, thus, the algorithm should only provide one coefficients as being 1 for each signature profile

library(MIXTURE)
# Load the TIL10 signature from Finotello et al.
data(TIL10)
# Signature Format
head(TIL10)
SelfTest(TIL10)

How to generate simulated mixture profiles from a molecular signature and test MIXTURE algorithm ?

# Run the MIXTURE on simulated samples built from the given signature
res <- SimulationTest(signatureMatrix = TIL10, 
                      maxCoefs = 6, #maximum number of cell types to include in the mixture (from 2 to maxCoefs)
                      maxSamples = 100, 
                      noisy = TRUE, #Add a noisy profile to the simulated pure mix
                      useCores=3L)
# Getting simulated data
# Simulated Samples
dim(res$SimulatedData$M)
head(res$SimulatedData$M[, 1:4])
#Simulated betas (coefficients)
head(res$SimulatedData$B)

#Retrieving MIXTURE results
dim(GetMixture(res$MIXTURE))

# Displaying the cell type proportions
p <- ProportionPlot(res$MIXTURE)
p
# Since ProportionPlot returns a ggplot object, we can change the characteristics of the plot beyond those relative to the bars and colors.
p + theme(axis.text.x = element_text(angle = 45, hjust = 1, vjust=0.5))

How to Download CDC1000 pure cell lines data to test the MIXTURE algorithm ?

library(data.table)
library(openxlsx)
library(org.Hs.eg.db)
library(limma)

#download and unzip expression from
url <- "https://www.cancerrxgene.org/gdsc1000/GDSC1000_WebResources/
/Data/preprocessed/Cell_line_RMA_proc_basalExp.txt.zip"
fname <- basename(url)
# this will save the downloaded file in your working directory
download.file(url  =  url, destfile = fname, method = "auto")
unzip(fname)

#load expression matrix
a.data<- fread("Cell_line_RMA_proc_basalExp.txt")

#update annotation
b.annot<- a.data[,1:2]
colnames(b.annot)<- c("symbol", "name")
columns(org.Hs.eg.db)
b.entrezids <- mapIds(org.Hs.eg.db, keys=b.annot$symbol, column="ENTREZID", 
                      keytype="SYMBOL", multiVals="first")
b.entrezids[sapply(b.entrezids, is.null)]<- NA
b.annot$entrezid<- unlist(b.entrezids)

#fix colnames
colnames(a.data)<- gsub("DATA.", "", colnames(a.data))

#make elist
b.elist<- new("EList", list(E=a.data[,-c(1,2)], genes= b.annot))
dim(b.elist)

#remove missing entrezid
b.elist<- b.elist[which(!is.na(b.elist$genes$entrezid)),]

#combine repeated entrezid expression
b.elist<- avereps(x = b.elist, ID = b.elist$genes$entrezid)
# this will save the cell lines gene expression matrix in your working directory
saveRDS(b.elist, file = "celllines.rds")

Once the file has been downloaded and stored in your hard disk, the you can proceed to analyze

# we will run over the first 10 cell lines
library(limma)
library(MIXTURE)
cells <- readRDS("celllines.rds")
# gene expression cell line data is in log2 scale
# so we anti log the data
M <- 2^cells$E 

cn <- colnames(cells$E)
cn[1:10]
rownames(M) <- cells$genes$symbol
# we will only process the first ten cell lines
cells.mix <- MIXTURE(expressionMatrix = M[, 1:10], signatureMatrix =  LM22, useCores = 3L)
head(GetMixture(cells.mix))

How to process TCGA Data ?

#Download FPKM data with TCGAbioLinks
library(TCGAbiolinks)
library(SummarizedExperiment)

query_all_brca_fpkm <- GDCquery(project = "TCGA-BRCA",
                             data.category = "Transcriptome Profiling",
                             data.type = "Gene Expression Quantification", 
                             experimental.strategy = "RNA-Seq",
                             workflow.type = "HTSeq - FPKM")

GDCdownload(query_all_brca_fpkm)
data_all_brca_fpkm <- GDCprepare(query_all_brca_fpkm, save = TRUE, save.filename = "BRCA.All.FPKM.rda")
##Chage the directory according to wahre you save your downloaded data

path <- "/home/elmer/Dropbox/IDEAS/cibersort/MIXTURE/"
library(TCGAbiolinks)
library(SummarizedExperiment)

```r

set path where your BRCA.All.FPKM.rda is located

load(file.path(path,"BRCA.All.FPKM.rda"))

# Prepare your gene expression data matrix gene_annot_all <- rowData(data) exp.all <- assay(data) target.data.all <- as.data.frame(colData(data)) tumor.type <- do.call(rbind,str_split(as.character(target.data.all$barcode),"-")) # Tissue type distribution table(tumor.type[,4]) rownames(exp.all)[1:10] # only ensembl gene ids exp.all <- exp.all[gene_annot_all$ensembl_gene_id,] #MIXTURE use rownames to identify genes and macth with the molecular signature rownames(exp.all) <- gene_annot_all$external_gene_name rownames(exp.all)[1:10] #an E list from limma package a.data <- new("EList",list(E = exp.all, genes = gene_annot_all, targets = target.data.all)) a.data$targets$TissueType <- do.call(rbind,str_split(as.character(target.data.all$barcode),"-"))[,4] # processing with MIXTURE brca.mix <- MIXTURE(expressionMatrix = a.data$E, signatureMatrix = LM22, useCores = 6L)

round(apply(GetMixture(brca.mix),2,function(x) sum(x>0))/nrow(GetMixture(brca.mix)),2) brca.prop <- GetMixture(brca.mix) df <- data.frame(Proportions = c(brca.prop), CellTypes = rep(colnames(brca.prop), each = nrow(brca.prop)))

ggplot(df,aes(x=CellTypes, y=Proportions, fill = CellTypes)) + geom_boxplot(outlier.size = 0.7) + theme(axis.text.x = element_text(angle=45, vjust = 1,hjust = 1)) ```

elmerfer/MIXTURE documentation built on Aug. 20, 2024, 8:03 p.m.