Transposable element (TE) expression in cancer is a well defined characteristic of the cancer transcriptome (source) however, TE expression is traditionally overlooked in RNA-seq analyses due to difficulties involving mappability, sequence similarity, and quality of annotations (source).
This repository contains the scripts needed for uniform processing of publicly available RNA-seq data from GEO with the explicit purpose of accurately quantifying gene and transposable element expression in cancer cell lines treated with epigenetic and cytotoxic drugs. This project has three main arms:
- A pipeline for downloading and processing (quantifying) gene and TE expression from publicly available data hosted on NCBI GEO. This custom pipeline implements a modification of the REDiscoverTE method which has been shown to accurately quantify transcript expression as well a TE expression at the loci level. One of the important modifications to this existing pipeline is the inclusion of decoy sequences to avoid spurious annotations as well a implementing proper offsets to transcript level counts when summarizing to the gene-level (source).
- The second arm of this repository is a uniform differential expression analysis which assesses
and attempts to control for violations of the assumptions made by global scaling normalization
methods. An important step in comparing expression values between conditions during differential
expression analysis is normalization. Violations of these assumptions (i.e. global expression
changes) can lead to spurious DE results. Here, I have implemented an automated detection for
violations of these global normalization assumptions using the quantro
R package. If violations of these assumptions are detected then normalization is performed using
smooth quantile normalization (qsmooth)
in order to preserve these global expression differences. Appropriately normalized data are then
tested for differential expression against a fold-change threshold using the limma::treat
method. These differential expression results, along with the fitted models, raw and normalized
counts, are then exported to
SummarizedExperimentobjects for each individual bioproject. - The last arm of this project involves collecting each uniformly processed bioproject's differential expression results into a single database back-end and exposing the data with a Shiny application. The Shiny application implements several novel meta-analysis and visualization functions for the exploration of large-scale patterns of differential expression across datasets. It is also worth mentioning that considerable effort was taken to standardize the cell type and drug annotations for all samples in this database. Cell types are given uniform labels from ATCC and drugs are assigned functional labels from medchemexpress when available. The code for the actual Shiny application implementation is located in a separate github repository. The meta-analysis application has been useful for hypothesis generation. For instance, observing drug treatments that result in expression patterns which deviate from other samples in their drug class has led to further studies regarding CDK12 inhibition.
- appdata/: contains all data used to create and used by the meta-app. scripts/appdata.R details the creation of the SummarizedExperiment object used as the backend for the database application. Since this data is fairly large it is not present in this repository.
- bioprojects/: contains subdirectories named by their NCBI bioproject ID from GEO. Each subdirectory contains all raw and processed data, analysis scripts, and analysis output for that particular bioproject. Currently, only the analysis scripts and annotation information are present in this repository. Like the appdata above, raw data is housed on the bioinformatics servers at Coriell.
- scripts/: contains all processing scripts used to download data from SRA and process for use in the meta-app. The scripts generally describe the processing steps for a single bioproject.