TCGA study case

Table of content

• Introduction
• Data sources
• Data preprocessing
• Use in the different courses

Introduction

This page describes a study case based on data from The Cancer Genome Atlas (TCGA; https://cancergenome.nih.gov/). This dataset contains more than 11,285 samples from patients suffering of a wide variety of cancer types. We use some subsets of this huge dataset for different courses of the Diplôme Universitaire en Bioinformatique Intégrative (DU-Bii).

The full datasets are available in the NCBI databases (Gene Expression Ombinus, Short Read Archives).

For the sake of simplicity, we took benefit of pre-processed data made available by Ron Shamir’s team.

We provide here

• the links to the sources of information,
• the scripts used to extract subsets of the original datasets, and format the them according to our particular needs.

Data sources

Publications

• G. Ciriello et al. (2015). Comprehensive Molecular Portraits of Invasive Lobular Breast Cancer, Cell. 163: 506–519. https://doi.org/10.1016/j.cell.2015.09.033.

• https://www.nature.com/articles/nature11412

TCGA web site

https://cancergenome.nih.gov/

Preprocessed datasets made available by Ron Shamir’s team

• Multi-Omics Cancer Benchmark TCGA Preprocessed Data
• Publication: https://academic.oup.com/nar/article/46/20/10546/5123392
• Data availability: http://acgt.cs.tau.ac.il/multi_omic_benchmark/download.html

• Clinical annotation for the samples: http://acgt.cs.tau.ac.il/multi_omic_benchmark/data/clinical.zip

• Note: the data is pre-normalised (library scaling + log2 transformation)

• Selected study case: Breast cancer dataset

  • Data tables: http://acgt.cs.tau.ac.il/multi_omic_benchmark/data/breast.zip

  • Relevant columns for classification: 2 markers used by clinicians to assign tissues to cancer subtypes: ER_Status_nature2012 and HER2_Final_Status_nature2012

  • Laura Cantini prepared an additional file with the subtypes (Human/subtypes_annotation.txt) + a script enabling to produce this subtype label based on the combination of these two markers (Human/SubAnnot.R)

Recount2

Data preprocessing

We downloaded the TCGA raw counts from the Recount2 database, and applied the following preprocessing steps:

1. select the samples belonging to the Breast Invasive Cancer (BIC) study;
2. define the cancer type (used as class label for supervised classification) based on the three immuno markers;
3. filter out “undetected” genes, i.e. genes having zero counts in almost all samples;
4. library size standardisation;
5. log2-transform of the counts;
6. detection of differentially expressed genes
7. selection of a reduced subset of genes (likely to be relevant for clustering and supervised classification) by keeping the 1000 genes having the lowest adjusted p-value in differential expression analysis;
8. exported the different results (raw counts, filtered, normalised, differentially expressed) in TSV files.

The preprocessing was done with an R markdown file, which enables anyone to reproduce the results and understand each step.

• Pre-processing report: import_TCGA_from_Recount.html
• R markdown: import_TCGA_from_Recount.Rmd

Use in the different courses

• Module 3 - Statistical analysis with R:
  • Differential analysis
  • Clustering
  • Evaluation of classifications: compare clustering results with markers
  • Supervised classification

• Module 6 - Integrative bioinformatics
  • Multi-level factorisation of multi-omics data (Laura Cantini)