This file contains instructions on how to use the code associated to the paper "Optimal miRNA sequencing depth to predict cancer patient survival".

How to download the TCGA data (optional, a sample with 100 patients is provided)

To overcome runing time and size issues, the data available in 'Rdata_cancer_screen/' contain only 100 patients for 3 cancers (ACC, BRCA and KIRC) in .RData files.

Clinical and mRNA-seq datasets were obtained using the Broad GDAC FIREHOSE utility.

1. Go on the website https://gdac.broadinstitute.org
2. Choose the cancer and click on 'Browse' in the 'Data' column
3. Clinical data: 3.a. On the 'Clinical' panel, click on 'Merge_Clinical'. A .zip file is downloaded. 3.b. Open this .zip file and copy the file 'cancer-name'.clin.merged.txt' (e.g. 'KIRC.clin.merged.txt') into the folder '../data_cancer/'cancer-name'/' (e.g. '../data_cancer/KIRC/').
4. mRNA data: 4.a. On the 'mRNASeq' panel, click on 'mRNAseq_Preprocess'. A zip file is downloaded. 4.b. Open this .zip file and copy the file ''cancer-name'.uncv2.mRNAseq_RSEM_all.txt' (e.g. 'KIRC.uncv2.mRNAseq_RSEM_all.txt') into the folder '../data_cancer/'cancer-name'/' (e.g. '../data_cancer/KIRC/').

In this script, we use RData files that we created to load RNA-seq and clinical data. To use TCGA data downloaded as mentionned above, please use the commented part of the file 'R_scripts/load_data/load_data_final.R'.

How to run the code

1. Open the file 'R_scripts/Survival_seq_depth.Rproj' to open RStudio environment.
2. Open the 'R_scripts/main.R' file and follow the instruction in the script.

As the running time is high, results will be saved all along the code in 'R_scripts/data_fit' into .RData files. Please set 'learn_new_models' variable to 'TRUE' if you want to learn new models. Otherwise, previous models (if they exist) will be loaded.

Running times

The article value for the number of models to be learned is 50 (10 repetitions of 5-fold cross-validation). It can take a lot of times with 8 percentage of patients in the training set, 6 degradation factors of the sequencing depth, and 50 models learned in each case (typically 4 hours for one cancer), and we advice the user to first set K_folds = 3 (3 folds) and n_rep = 2 (2 repetitions) for a first check.