Steps to reproduce this analysis

This is a step-by-step guide on how to reproduce exactly the analysis performed in the paper.

Note #1: To run all steps, you can type bash run_all_steps.sh at the command line while inside This directory. This script presumes that you've already run Fatiscan and FatiGO. If you do not have the fatiscan_output and fatigo_output directories, the script currently breaks at step 7 or step 9, respectively.

Note #2: Each bash script for each step assumes that they are run from their directory. The scripts will break otherwise.

Note #3: The annotations and the TCGA data are not included with this repository because of their size (several GB total). This pipeline assumes that you do step 1a and 1b.

Step 0

install necessary R packages (DESeq2, ProMISe) and python packages (BCBio and HTSeq)

• run bash step_0.sh to install the necessary packages (use sudo bash step_0.sh to install system-wide)
  • this presumes that R is already installed: to run this particular analysis, you need to install R-3.1.3
  • If you need to install it, follow these instructions:
    • download R-3.1.3 from here: https://cran.r-project.org/src/base/R-3/R-3.1.3.tar.gz
    • run tar -xzf R-3.1.3.tar.gz in the directory with the downloaded tarball.
    • navigate into the R-3.1.3 directory, and then run either of the following set of commands:
      • if you have no other versions of R, run: ./configure and then make, then make install
      • if you have other versions of R, run ./configure --prefix=/some/other/path before running make and make install
    • be sure to use the right path to Rscript in the following steps

Note: you may need to change the top line of the install_packages.R script to reflect the proper path to Rscript. If there is an error, you can check the path by typing which Rscript at the command line.

Note: you will need to change the directory paths at the beginning of each bash script currently to run this on your own machine.

Step 1a

Download all samples with clinical data, RNA-SeqV2, and miRNA-Seq data available.

1. LUAD: 441 (19 control samples, 422 tumor samples)
   • use luad_file_manifest.txt with the GDC data transfer tool to download the same files yourself
2. LUSC: 367 (37 control samples, 330 tumor sample)
   • use lusc_file_manifest.txt with the GDC data transfer tool to download the same files yourself

The easiest way to do this is to follow the steps in the tcga_file_manifests directory using the GDC data transfer tool. Click here to follow those steps.

If you download the data manually, please keep in mind the following note:

NOTE: For the downstream steps to work without modification, you must rename the top level directories for this data as LUAD_data and LUSC_data, respectively, and place these directories in this repository. If you wish to use different names, or place directories elsewhere on your computer, you need to modify the LUAD_DIR and LUSC_DIR variables at the top of the following scripts: step_2c.sh, step_3.sh, step_4.sh, step_5.sh, step_7.sh, and step_9.sh

Step 1b

Download annotations used for TCGA analysis

• RNA: UCSC, hg19, June 2011; miRNA: miRBase v21 and miRanda 08/2010 release
• run step_1b.sh to do two things:
  • download UCSC tables, the miRNA annotations and initial miRanda miRNA-target predictions
  • generate the table that matches pathway IDs to intelligible pathway names NOTE: this download takes up approximately 880MB of space

Step 2

Prep the UCSC annotations to facilitate downstream analyses

• run step_2a.sh to prepare the UCSC annotations to facilitate downstream analyses Output: a combined kgXref table to facilitate mapping UCSC IDs to gene names

Step 2b

Prep the miRNA annotations to facilitate downstream analyses

• run step_2b.sh Outputs:
  • a list of all miRNAs (and accessions) for both the target matrix and miRBase v21
  • a list of all targets
  • a matrix of interactions (1 if yes, 0 if no)

Step 2c

Prep the TCGA input files for downstream analyses

• run step_2c.sh
• this automatically runs on both LUAD and LUSCS directories Outputs:
  • miRNA count matrix ready for step 3 is in (LUAD|LUSC)_data directory
  • miRNA counts ready for 4 are in (LUAD|LUSC)_data/miRNASeq/BCGSC__50/Level_3/compressed_miRNAcounts directory
  • RNA normalized isoform counts ready for steps 3 and 4 are in (LUAD|LUSC)_data/RNASeqV2/UNC__58/Level_3/TCGA_isoform_normalized_results directory

Step 3

Run DESeq on the miRNA expression values, as well as the RNA isoform values

• run step_3.sh Output: DESeq2 results are in (LUAD|LUSC)_data/deseq_results directories. The *Rank.txt file is the one used in step 6 for the Fatiscan analysis.

Step 4

Run ProMISe

• run step_4.sh Output: the promise RData will be stored in a directory called output

Step 5

Identify updated miRNA-mRNA predictions for each dataset

• run step_5.sh to process the ProMISe results Output: this produces a GMT gene sets file with transcripts, and an extended gene set file. The latter is used for step 6, along with the rank file generated by step 3.

Step 6

Run Babelomics Fatiscan on DESeq2 results of using promise predictions

This requires manual manipulation of files

1. Make an account on Babelomics v4 (http://v4.babelomics.org)
2. Upload the extended miRNA-target gene set file with data type Annotation > Extended Annotation
3. Upload the DESeq2 rank list of transcripts with data type ID List > Ranked
4. Choose Functional Analysis, then Gene Set Analysis under Set Enrichment Analysis
5. Choose the rank list for input data; choose fatiscan for test, and choose two-tailed
6. choose your own annotations and select the extended list

Precomputed results from Fatiscan are found in the fatiscan_output directory.

Step 7

Run post-Babelomics processing on the results

1. first download the results from the Babelomics website. On the bottom of the results page, click Download Job
2. move the resulting job folder to the (LUAD|LUSC)_data directory and rename it fatiscan_output
3. run step_7.sh Output: a post_babelomics_results directory with processed results for up-regulated and down-regulated genes separately
   • in these directories (specifically, the downregulatedGenes/up-regulated_miRNAs/ and upregulatedGenes/down-regulated_miRNAs), you'll find *FatigoInput.txt lists of the genes we'll use as input for FatiGO in step 8.

Step 8

Run Babelomics FatiGO on gene-level DESeq2 results

Like step 6, this also requires manual input of data

1. Upload the up-regulated and down-regulated list of mRNA targets as ID List > Gene
2. Choose Functional Analysis then FatiGO under Single Enrichment Analysis
3. Choose Id list vs Rest of Genome
4. Choose the up- or down-regulated list (run separate analysis for each)
5. Remove duplicates? should be set to Remove from list2 those appearing in list1 (complementary list)
6. For databases, choose human for organism, and check boxes for GO biological process, GO cellular component, GO molecular function, KEGG pathways, Reactome, and Biocarta. Note: For this step, if you do both LUSC and LUAD, you'll run four separate jobs.

Precomputed results from this step are found in the fatigo_output directory

Step 9

Generate the Hive Plots

1. first download the results from the Babelomics website. Like before, you click on Download job
2. in the (LUAD|LUSC)_data directory, create a directory called fatigo_out
3. move the resulting job folder to the fatigo_out and rename it either upGene or downGene, depending on whether the FatiGO job focused on up-regulated or down-regulated genes.
4. run step_9.sh Output: PDF files of the hive plots.

Note: You'll have to manually add in annotations afterward using the tables and Adobe, Preview, or Powerpoint.

Note 2: If there are any issues with reproducing this analysis, please submit an issue on Github.