Evidence of antagonistic predictive effects of miRNAs in breast cancer cohorts through data-driven networks

Principal aim

In this repository we provide the codes necessary to replicate the findings presented in the research article Evidence of antagonistic predictive effects of miRNAs in breast cancer cohorts through data-driven networks authored by Cesare Miglioli, Gaetan Bakalli, Samuel Orso, Mucyo Karemera, Roberto Molinari, Stephane Guerrier and Nabil Mili.

The statistical analysis performed in this study is based on the data presented in the paper Subtype-specific micro-RNA expression signatures in breast cancer progression by Haakensen et al in 2016. We thank the authors for having made available the AHUS data set on the free access ArrayExpress platform.

Get the data from ArrayExpress in R

First install the ArrayExpress package, through bioconductor, with the following code:

### R version 3.5 and older ###

if (!requireNamespace("BiocManager", quietly = TRUE))
  install.packages("BiocManager")

BiocManager::install("ArrayExpress", version = "3.8")

### R version 3.6 ###

if (!requireNamespace("BiocManager", quietly = TRUE))
  install.packages("BiocManager")

BiocManager::install("ArrayExpress")


require(ArrayExpress) #load the new package

Now you can create a temporary directory to store the raw data listed as E-MTAB-3759 on ArrayExpress. Then you just need to extract the data thanks to the function ae2bioc() of the package. To have a first impression of the AHUS dataset, you can click on the following Visual Impression.

temp_dir <- "C:/..." #choose your preferred path

dir.create(temp_dir)
setwd(temp_dir)

# Download the raw files
# To run only the first time!

ae_obj <- ArrayExpress::getAE('E-MTAB-3759', type = 'full', path = temp_dir)

str(ae_obj) #to explore the new object

### Extraction ###

mtab3759raw <- ae2bioc(mageFiles = ae_obj)

The response y (i.e. if the subject has breast cancer or not) can be obtained directly from the mtab3759raw object together with the type of breast cancer y_sub (i.e. either benign, DCIS or invasive) which indicates the sub-populations. However, in order to get the final design matrix X of miRNAs used in the study, we need to apply the function normalizeBetweenArrays() of the limma package to our previously found object. This function normalizes expression intensities so that the same intensities (or log-ratios) have similar distributions across a set of arrays.

# Response variable

y <- mtab3759raw@phenoData@data$Factor.Value.disease.

y_sub <- mtab3759raw@phenoData@data$Characteristics.clinical.history. 

# Design matrix X of miRNAs

mtab3759raw <- normalizeBetweenArrays(mtab3759raw,method="quantile")

X <- mtab3759raw@.Data[[1]]

X <-  t(X) #to have subjects on the rows and miRNAs on the columns

colnames(X) <- mtab3759raw$genes$GeneName #Assign gene names to each column

Reproduce the results

Having now the final design matrix X and the response y, we are able to reproduce the results and the graphs contained in the research article. Please refer to the file breast_cancer_data_analysis.R in this repository for a detailed explanation.