AI summary by TwinMind
StatQuest: A gentle introduction to RNA-seq
StatQuest with Josh Starmer
18m 25s
Introduction & Overview
- StatQuest tutorial on RNA-seq methodology
- Sponsored by UNC Chapel Hill Genetics Department
- Goal: Compare normal vs mutated neural cells to identify genetic mechanisms causing behavioral differences
- Focus: Differences in gene expression between cell types
RNA-Seq Concept & Purpose
- High-throughput sequencing identifies which genes are active and transcription levels
- Measures gene expression in normal cells vs mutated cells
- Compares two cell types to identify differences
- Example outcomes:
- Gene 1: No difference between normal/mutated
- Gene 2: Big difference between normal/mutated
- Gene 3: Subtle difference between normal/mutated
Three Main RNA-Seq Steps
- Prepare sequencing library
- Sequence
- Data analysis
Library Preparation (Illumina Protocol)
- Step 1: Isolate RNA from cells
- Step 2: Fragment RNA into small pieces
- RNA transcripts can be thousands of bases long
- Sequencing machines only handle 200-300 base pair fragments
- Step 3: Convert RNA fragments to double-stranded DNA
- DNA more stable than RNA
- Easily amplified and modified
- Step 4: Add sequencing adapters
- Allow sequencing machine to recognize fragments
- Enable multiplexing (different samples with different adapters)
- Saves time and money
- Not 100% efficient - some fragments don't get adapters
- Step 5: PCR amplify library
- Only fragments with adapters are amplified
- Enriches properly prepared fragments
- Step 6: Quality control
- Verify library concentration
- Verify fragment lengths (not too long/short)
Sequencing Process
- Flow cell setup: ~400 million DNA fragments arranged vertically in grid
- Sequencing mechanism:
- Fluorescent probes color-coded by nucleotide type
- Probes attach to first base in each sequence
- Machine photographs flow cell from above
- Identifies base by color (A, G, C, T)
- Washes color off probes
- Repeats for next base position
- Continues until complete sequence determined
Sequencing Quality Issues
- Low confidence calls: Probes don't shine bright enough
- Low diversity: Too many same-color probes in region
- Colors blur together
- Especially problematic in first few nucleotides
- Affects machine's ability to locate fragments on grid
- Quality scores: Reflect machine confidence in base calls
Raw Data Format
- Each sequencing read = 4 lines of data:
- Unique ID (starts with @ symbol)
- Base sequence of fragment
- Plus character (+) - purpose unknown
- Quality scores for each base
- Typical run: 400 million reads = 1.6 billion lines of data
Data Processing Steps
Filter Garbage Reads
- Low quality base calls: Poor quality scores
- Chemistry artifacts: Adapter-only reads (adapters bound to each other)
Align Reads to Genome
- Start with reference genome sequence
- Split genome into small fragments
- Create index of fragments and their genome locations
- Split sequencing reads into fragments
- Match read fragments to genome fragments
- Determine chromosome position for each read
- Fragmentation benefit: Allows alignment even with inexact matches (accounts for individual genome differences)
Count Reads Per Gene
- Use chromosome position to determine if read falls within gene coordinates
- Human genome: ~20,000 genes
- Results in count matrix: genes × samples
Data Matrix Structure
- Rows: ~20,000 genes (gene names in first column)
- Columns: Sample counts
- Bulk RNA-seq: 6-8+ samples (average of ~6 million cells per sample)
- Example: 3 normal + 3 disease = 6 total samples
- Single-cell RNA-seq: Each cell = individual sample (800+ samples possible)
- Creates huge matrix: 20,000 × 800+ = 16+ million values
Data Normalization
- Problem: Different total read counts per sample due to:
- Varying low-quality read proportions
- Different flow cell concentrations
- Example: Sample 1 (635 reads) vs Sample 2 (1,270 reads)
- Solution: Adjust read counts to reflect total reads per sample
- Simple method: Divide gene counts by total mapped reads per sample
- Note: More sophisticated normalization methods available
Data Analysis
Step 1: Plot the Data
- Challenge: 20,000 genes = 20,000 axes needed
- Solution: Use PCA (Principal Component Analysis)
- Reduces dimensions while preserving important data aspects
- Most important differences on x-axis
- Less important differences on y-axis
- Bulk RNA-seq PCA example:
- WT (normal) samples: Left side, spread on y-axis
- KO (mutated) samples: Clustered in corner
- Biggest differences between WT and KO groups
- Single-cell RNA-seq PCA example:
- Green cells: Stationary
- Orange cells: Mobile
- Most orange cells separated from green
- Some orange cells similar to green (may exclude from analysis)
- PCA benefits:
- Identifies expected interesting differences
- Reveals samples to potentially exclude from downstream analysis
Step 2: Identify Differentially Expressed Genes
- Tools: R programming language with edgeR or DESeq2
- Visualization: MA plot or volcano plot
- Red dots: Significantly different genes between normal/mutant
- Black dots: No significant difference
- X-axis: Gene transcription level (CPM - counts per million)
- Left: Lowly transcribed genes
- Right: Highly transcribed genes
- Y-axis: Magnitude of difference (log FC - log fold change)
Step 3: Interpretation
- Known hypothesis: Validate if experiment supports expectations
- Exploratory analysis: Identify enriched pathways in normal vs mutant gene sets
- Further analysis: Requires additional statistical methods (PCA, heatmaps, p-values, false discovery rates)
