Chapter 12: Transcriptomics: Analyzing Gene Expression (RNA-Seq)

📥 Download Lecture Slides (PPTX)

12.1 What is the Transcriptome?

The genome is the cell's complete set of DNA—a static blueprint. However, not all genes in the blueprint are active at the same time. A brain cell and a liver cell have the same DNA, but they use different sets of genes to perform their specialized functions.

The transcriptome is the complete set of all RNA molecules (transcripts) in a cell at a specific moment. It's a dynamic snapshot of which genes are currently being expressed.

Transcriptomics is the study of this transcriptome, primarily to understand which genes are turned "on" or "off" under different conditions.

12.2 The RNA-Seq Workflow

The dominant technology for studying the transcriptome is RNA-Seq (RNA Sequencing). It leverages the NGS technology we learned about in Chapter 11.

1. RNA Extraction: Isolate all RNA from a cell sample.
2. Reverse Transcription: Convert the unstable RNA into more stable complementary DNA (cDNA).
3. NGS Sequencing: Sequence the cDNA fragments to generate millions of short reads.
4. Mapping: Align the sequencing reads to a reference genome to see which gene they came from.
5. Quantification: Count how many reads mapped to each gene. This count is a proxy for the gene's expression level.

12.3 Mapping Reads to a Genome

Unlike genome assembly where we build the puzzle from scratch, in RNA-Seq we usually have a reference genome to work with. We use specialized aligners to map our RNA reads back to it.

However, there's a complication: splicing. In eukaryotes, genes are composed of exons (coding regions) and introns (non-coding regions). The introns are "spliced out" of the final mRNA. This means a read from a single mRNA molecule might map to two different exons in the genome, separated by a large intron.

Therefore, we need splice-aware aligners like STAR or HISAT2 that can handle these large gaps.

12.4 Quantifying and Normalizing

Simply counting the reads per gene is not enough. A longer gene will naturally get more reads than a shorter one, and a sample sequenced to a greater depth will have more reads overall. We must normalize these counts.

Common units include: * TPM (Transcripts Per Million): Normalizes for both gene length and sequencing depth. This is the most common unit for comparing expression levels within a sample. * FPKM/RPKM: Older metrics, largely replaced by TPM.

12.5 Differential Expression Analysis

The most common goal of an RNA-Seq experiment is to find Differentially Expressed Genes (DEGs).

• Scenario: You have two conditions: "Treated" vs "Control".
• Question: Which genes are significantly up-regulated (more active) or down-regulated (less active) in the "Treated" group?

We use statistical tools like DESeq2 or edgeR that take the raw counts, perform robust normalization, and calculate two key values for each gene:

1. Log2 Fold Change: How much the gene's expression has changed between conditions. (e.g., +2 means 4x higher expression).
2. p-value (or adjusted p-value): The statistical significance of this change.

The final result is often visualized as a Volcano Plot, which plots the fold change against the p-value, making it easy to see the most significant DEGs.

12.6 Bioinformatics in Action: A Conceptual Workflow

A full RNA-Seq analysis involves multiple command-line tools chained together. While the exact code is too complex for an introduction, the conceptual pipeline looks like this:

# Step 1: Index the reference genome for the aligner
hisat2-build genome.fa genome_index

# Step 2: Align reads from a sample to the genome
# -x: index, -U: unpaired reads, -S: output file
hisat2 -x genome_index -U sample1.fastq -S sample1.sam

# Step 3: Convert SAM to sorted BAM (a compressed, indexed format)
samtools view -bS sample1.sam | samtools sort -o sample1.sorted.bam

# Step 4: Count how many reads overlap with each gene
# -a: annotation file, -o: output counts
featureCounts -a genes.gtf -o counts.txt sample1.sorted.bam

# Step 5: Analyze the 'counts.txt' file in R using DESeq2
# (This step is performed in the R programming language, not the shell)

This workflow is repeated for every sample, and the final count files are combined for differential expression analysis.

Summary

Transcriptomics provides a dynamic view of the genome in action. The RNA-Seq workflow allows us to quantify gene expression by sequencing cDNA and mapping the reads back to a genome. The ultimate goal is often to find differentially expressed genes between different conditions, giving us powerful insights into the molecular basis of disease, development, and cellular responses.

12.7 Modern Quantification Methods: Pseudoalignment and Transcript-Level Analysis

Newer tools perform very fast, accurate transcript quantification without full alignment. These are now standard for many RNA-Seq workflows:

• Pseudoaligners / lightweight quantifiers: Salmon and Kallisto map reads to a transcriptome index and produce transcript-level abundances quickly and with low memory.
• Transcript-to-gene aggregation: Use tximport (R) to import transcript-level estimates into gene-level counts for DE testing in DESeq2 or edgeR.

Example: quantify with salmon (quasi-mapping mode):

# Build transcriptome index
salmon index -t transcripts.fa -i transcripts_index

# Quantify reads
salmon quant -i transcripts_index -l A \
  -1 sample_R1.fastq.gz -2 sample_R2.fastq.gz \
  -p 8 -o sample_salmon

12.8 Differential Expression and Downstream Analysis

• DE tools: DESeq2 and edgeR remain standard for differential expression with appropriate normalization and experimental design specification.
• Visualization: PCA on normalized counts, heatmaps, and volcano plots help inspect results and batch effects.
• Single-cell: For single-cell RNA-Seq, specialized preprocessing is required: UMI handling, cell barcode parsing, doublet detection, and normalization. Tools: CellRanger (10x Genomics), Scanpy, Seurat.

12.9 Reproducible RNA-Seq Pipelines

Encode your pipelines using Snakemake or Nextflow, containerize tools, and include MultiQC to aggregate QC across samples. For large consortia projects, maintain metadata (sample sheets) and workflow parameter files to ensure provenance.