Bulk RNA-seq Differential Expression Across Human Tissues

This repository contains a cleaned and reproducible R workflow for comparing bulk RNA-seq expression profiles from brain, lung, and skin GTEx samples. The analysis uses recount/recount3 data structures and edgeR for filtering, TMM normalisation, multidimensional scaling, quasi-likelihood differential expression, tissue-specific gene identification, and marker validation.

Main analyses

• Reconstruction of read counts from recount objects.
• Calculation of TPM and CPM expression values.
• Sample-level QC using RIN, library size, mapping rate, rRNA fraction, and mitochondrial read percentage.
• Removal of mitochondrial/non-canonical chromosomes and short-RNA features.
• Selection of three manually reviewed samples per tissue.
• Low-expression filtering with filterByExpr.
• TMM library-size normalisation.
• MDS evaluation of tissue separation and possible QC-related effects.
• Pairwise edgeR quasi-likelihood tests:
  • Lung vs Brain
  • Skin vs Brain
  • Skin vs Lung
• Identification of genes consistently upregulated in each tissue relative to both other tissues.
• TPM-based validation of TTBK1, VSTM2B, TBX4, and HOXC10.

Data

The RDS files are not included because they can be large and may be subject to course or data-distribution restrictions.

Installation

From the repository root, run:

Rscript install_packages.R

Run the analysis

Rscript analysis/run_analysis.R

The selected sample indices and statistical thresholds can be changed in config/config.R.

Default thresholds

• FDR: < 0.01
• Absolute log2 fold change: > 1

A gene is labelled tissue-specific and upregulated only when it passes both thresholds in comparisons with each of the other two tissues.

Important improvements over the original script

• Removed repeated package installation and repeated TPM calculations.
• Replaced absolute Windows paths with project-relative paths.
• Consolidated repeated brain/lung/skin processing into reusable functions.
• Replaced hard-coded gene row numbers with gene-name lookup.
• Kept continuous QC variables numeric instead of converting them to factors.
• Removed duplicated MDS code and empty blocks.
• Corrected extraction and ranking of edgeR result tables.
• Added automatic output folders, descriptive file names, and session details.
• Preserved the original script under legacy/ for full traceability.

Reproducibility note

The sample choices in config/config.R reproduce the manual selections in the original analysis. They should be re-evaluated if different recount objects or sample orders are used.