Explain whole-genome and exome sequencing

What Are We Actually Sequencing?

WHY it matters: Captures structural variants (large deletions, duplications, inversions), regulatory mutations in promoters/enhancers, non-coding RNA genes, and mitochondrial DNA—things exome sequencing mises.

WHY it matters: Cost-effective for diagnosing Mendelian disorders. If a mutation changes a protein's amino acid sequence (missense, nonsense, frameshift), exome sequencing will catch it. Much cheaper than WGS (~$500 vs$1000-3000), and easier to interpret (fewer variants to sift through).

How Does Exome Capture Work? (Derivation from First Principles)

The Challenge: DNA fragmentation during Next-Gen Sequencing (NGS) breaks the genome into random ~200-400bp pieces. How do we fish out only the exon fragments?

Solution: Hybridization capture using biotinylated RNA or DNA probes.

Step-by-Step Mechanism:

1. Fragment the genome → Extract genomic DNA, shear it (sonication or enzymatic) into200-400bp fragments
2. Add adapters → Ligate sequencing adapters (with barcodes/indices) to fragment ends
3. Hybridization → Mix fragments with a library of ~300,000-400,000 biotinylated oligonucleotide probes (120-mer sequences) that are complementary to all known exon sequences
   • WHY biotin? Biotin binds tightly to streptavidin. This is the "molecular Velcro" we'll use for capture.
4. Capture → Add streptavidin-coated magnetic beads. The biotin-probe-exon complexes stick to the beads.
5. Wash away → Magnetic separation removes all non-exonic fragments (they weren't bound by probes, so they wash away)
6. Elution → Heat denatures the probe-DNA hybrids, releasing captured exon fragments
7. Sequence → The enriched exonic library goes into the NGS sequencer (Illumina, etc.)

$E = \frac{(\text{exon reads} / \text{total reads})_{\text{after capture}}}{(\text{exon reads} / \text{total reads})_{\text{before capture}}}$

Derivation of expected coverage:

• Human exome size: $L_{\text{exome}} \approx 30 \times 10^6$ bp
• Human genome size: $L_{\text{genome}} \approx 3.2 \times 10^9$ bp
• Fraction of genome that is exome: $f = \frac{30 \times 10^6}{3.2 \times 10^9} \approx 0.009$ (0.9%)

If we generate $N$ total sequencing reads of length $\ell$ (say150bp):

• Without capture: Expected exome coverage depth = $\frac{N \cdot \ell \cdot f}{L_{\text{exome}}}$
• With capture at efficiency $\eta$ (fraction of reads that are on-target, typically 0.7-0.9):

$\text{Exome coverage} = \frac{N \cdot \ell \cdot \eta}{L_{\text{exome}}}$

Example: 100 million reads × 150bp × 0.8 on-target efficiency / 30Mbp = 40× mean coverage of exons.

Approach: Order exome sequencing.

Result: Identified a de novo missense variant in SCN2A gene (encodes a voltage-gated sodium channel). The mutation was c.5645G>A (p.Arg1882Gln) in exon 28.

WHY exome worked here:

1. Single-gene disorder → the causal variant is in a coding region
2. De novo → not inherited, so parents' exomes are normal (trio analysis helps filter)
3. SCN2A is a known epilepsy gene → variant interpretation is straightforward

WHEN exome would FAIL: If the patient had a deep intronic mutation affecting splicing (e.g., in SCN2A intron 10), WES wouldn't sequence that region. WGS would catch it.

WGS reveals: A 350kb deletion on chromosome 16p11.2 (a known autism-associated CNV). This deletion removes multiple genes entirely.

WHY WES missed it:

• Exome sequencing uses read-depth analysis to infer deletions (if exons in a gene have50% normal coverage, suspect heterozygous deletion).
• BUT: If capture probes don't bind well, or the deletion is in a gene not well-covered, WES has low sensitivity.
• WGS gives uniform coverage across deleted and flanking regions → structural variant calers (Manta, Delly) detect the breakpoints by finding discordant read pairs and split reads.

The math:

• WES coverage is non-uniform (10× in some exons, 100× in others, 0× in introns).
• WGS coverage is ~30-40× everywhere → statistical power to detect a50% drop (from 40× to 20×) is much higher.

Whole-Genome Sequencing: What Extra Do We Get?

$P(X \geq k) = 1 - \sum_{i=0}^{k-1} \frac{C^i e^{-C}}{i!}$

WHY Poisson? Sequencing reads are generated randomly across the genome. If mean coverage is $C$, the number of reads covering any specific base follows a Poisson($C$) distribution.

Example: At 30× mean coverage, what fraction of bases are covered ≥10×?

• $P(X \geq 10)$ when $C=30$: We compute $1 - P(X \leq 9) = 1 - 0.0003 \approx 99.997\%$

For variant calling:

• SNVs require ~10-15× coverage for confident heterozygote calls
• Indels require ~20× for reliable detection
• Structural variants need ~30× for breakpoint resolution

What WGS detects that WES misses:

1. Regulatory variants: Mutations in promoters (e.g., TERT promoter mutations in melanoma, glioma)
2. Deep intronic variants: Create cryptic splice sites (e.g., NF1 intron 31 variant causing neurofibromatosis)
3. Non-coding RNA genes: lncRNAs, miRNAs (e.g., MIR96 mutations cause hearing loss)
4. Repeat expansions: Huntington's disease (CAG repeats in HTT), Fragile X (CGG repeats in FMR1)—though short-read WGS struggles; long-read sequencing is better
5. Balanced translocations, inversions: Rearrangements that don't change copy number

WHY WES mises it: The mutation is 124bp upstream of the TERT start codon, in the promoter (non-coding). Standard exome capture probes don't target promoters.

Clinical impact: 60-80% of glioblastomas have these mutations. They're prognostic markers and potential therapeutic targets.

Comparison Table: When to Use Which?

Bioinformatics Pipeline (Simplified)

Both WGS and WES use similar analysis pipelines:

1. Base calling → Convert raw images (Illumina) to FASTQ (sequence + quality scores)
2. Alignment → Map reads to reference genome (hg38) using BWA-MEM → BAM file
3. Variant calling → Identify differences from reference:
   • SNVs/indels: GATK HaplotypeCaller, FreBayes
   • CNVs: CNVkit (WES), Manta (WGS)
   • SVs: Delly, Lumpy (WGS only)
4. Annotation → Predict variant effects (VEP, ANNOVAR): synonymous, missense, frameshift, splice-site
5. Filtering → Remove common variants (>1% frequency in gnomAD), low-quality calls
6. Interpretation → Match remaining variants to patient phenotype (ACMG criteria: pathogenic, likely pathogenic, VUS, benign)

Filtering cascade:

1. Population frequency: Remove if allele frequency $f > 0.01$ in gnomAD (unlikely to cause rare disease)
2. Predicted deleteriousness: Keep if CADD score $> 20$ (top 1% most deleterious), orPolyPhen/SIFT predict "damaging"
3. Inheritance pattern: If recessive disease suspected, keep only homozygous or compound heterozygous variants
4. Gene-disease association: Keep variants in genes known to cause the patient's phenotype (HPO term matching)

Math: If we have 25,000 variants:

• After frequency filter (keep rare): ~500 variants (98% removed)
• After deleteriousness filter: ~100 variants (80% removed)
• After inheritance filter (e.g., homozygous recessive): ~5-10 variants
• After gene-phenotype matching: 0-3 strong candidates

Why It Feels Right: Exome = protein-coding genes, diseases are caused by broken proteins, so exome should catch everything.

The Steel-Man (Why This Idea Has Merit): It's true that ~85% of known Mendelian disease mutations are in exons. So exome sequencing has high sensitivity for classic genetic diseases.

The Reality Check:

1. Coverage gaps: ~2-5% of exons are poorly captured (GC-rich regions, pseudogenes with high homology). If the causal variant is in a gap, WES mises it.
2. Non-coding diseases: Promoter mutations (TERT, FOXP2), enhancer mutations (limb malformation, β-thalassemia), intronic splicing mutations (deep intronic NF1, ABCA4 in Stargardt disease).
3. Structural variants: Large deletions, duplications, inversions are hard to detect with short reads and non-uniform coverage.
4. Repeat expansions: Huntington's, Fragile X, spinocerebellar ataxias—WES doesn't sequence these regions well.

The Fix: If WES is negative but clinical suspicion remains high:

• Consider WGS (adds 5-10% diagnostic yield)
• Targeted repeat expansion testing (Huntington's panel, Fragile X PCR)
• Mitochondrial genome sequencing (if myopathy/encephalopathy)
• RNA sequencing (detects aberant splicing from intronic variants)

Exome sequencing is like photocopying just those 20,000 important chapters (about 1% of the book). It's fast and cheap because you're not copying the whole book. If there's a typo in a chapter (a mutation in a gene), the photocopy will show it, and doctors can figure out why you're sick.

Whole-genome sequencing is like photocopying the ENTIRE book—all 3 billion letters, including the blank pages, the table of contents, the index, and the weird decorative margins. It costs more and creates a huge pile of paper, but now you can see if someone wrote secret notes in the margins (regulatory mutations) or ripped out whole pages (big deletions).

When do you need the whole book? If the doctor photocopied just the chapters (exome) and didn't find the typo, maybepo is in the table of contents or a margin note—something that controls WHEN a chapter gets read, not the chapter itself. Then you need the whole-genome sequencing to find it.

Connections

• Next-Generation Sequencing Technologies – the underlying Illumina/Ion Torrent tech
• Variant Calling and Annotation – bioinformatics pipeline details
• Copy Number Variation Detection – how CNVs are found differently in WGS vs WES
• RNA Sequencing – complementary approach to detect splicing defects missed by DNA sequencing
• Long-Read Sequencing – PacBio/Nanopore for repeat expansions and phasing
• Trio Analysis in Clinical Genetics – comparing proband + parents to filter de novo variants
• ACMG Variant Interpretation Guidelines – how to classify variants as pathogenic
• Pharmacogenomics – using WGS to predict drug metabolism
• Population Genomics – gnomAD database for filtering common variants

#flashcards/biology

What is whole-genome sequencing (WGS)? :: Determining the complete DNA sequence of an organism's genome (~3.2 billion bp in humans), including all coding and non-coding regions, regulatory elements, and structural variants.

Concept Map

Hinglish (regional understanding)