Describe next-generation sequencing (NGS)

Overview

Next-Generation Sequencing (NGS) revolutionized genomics by enabling massively parallel DNA sequencing—millions of fragments simultaneously, versus Sanger's one-at-a-time approach. This paralelism drops cost per genome from ~$100M (2001) to <$1000 (2020s) and accelerates projects from years to days.

[!intuition] The Core Idea

Imagine reading a book by:

1. Shreding it into millions of overlapping snippets
2. Reading all snippets at once (not one page after another)
3. Reconstructing the original text by finding overlaps

Sanger sequencing = reading one sentence at a time with a magnifying glass.
NGS = projecting all sentences onto a wall simultaneously and photographing them.

Why parallel? DNA polymerase (the "reader") works at ~1000 bp/s. Paralelizing across millions of clusters bypasses this biological speed limit—like having millions of photocopiers versus one.

[!definition] What is NGS?

A family of sequencing technologies that generate millions to billions of short reads (50–300 bp) simultaneously by:

1. Library preparation: Fragmenting DNA + adding adapters
2. Clonal amplification: Creating clusters of identical fragments
3. Sequencing-by-synthesis: Detecting nucleotide incorporation in real-time across all clusters
4. Bioinformatics: Aligning reads to a reference or assembling de novo

Key principle: Spatial separation + optical detection. Each DNA cluster occupies a unique location on a chip/bead; cameras record fluorescence signals across millions of sites per imaging cycle.

[!formula] Sequencing Depth and Coverage

1. Coverage Depth

$C = \frac{N \cdot L}{G}$ Where:

• $C$ = average coverage (how many reads overlap each genomic position)
• $N$ = total number of reads
• $L$ = read length (bp)
• $G$ = genome size (bp)

Derivation from first principles:

• Total bases sequenced = $N \cdot L$ (reads × length/read)
• These bases are randomly distributed across genome of size $G$
• Average times any position is read = $\frac{\text{total bases}}{G}$

Example: Human genome ($G = 3 \times 10^9$ bp), 100M reads ($N = 10^8$), 150 bp reads ($L = 150$): $C = \frac{10^8 \cdot 150}{3 \times 10^9} = \frac{1.5 \times 10^{10}}{3 \times 10^9} = 5\text{×}$ Why this matters: 30× is recommended for variant calling (ensures ~99.9% of genome covered≥5 times, critical for distinguishing true SNPs from errors).

2. Lander-Waterman Coverage Probability

For random sequencing, probability a base is covered≥1 time: $P(\text{covered}) = 1 - e^{-C}$

Derivation:

• Assume reads randomly placed (Poisson process)
• Probability a specific base is not hit by one read = $1 - \frac{L}{G}$
• For $N$ independent reads: $P(\text{not covered}) = \left(1 - \frac{L}{G}\right)^N$
• For large $G$, $\left(1 - \frac{L}{G}\right)^N \approx e^{-NL/G} = e^{-C}$
• So $P(\text{covered}) = 1 - e^{-C}$

Example: At 5× coverage: $P(\text{covered}) = 1 - e^{-5} = 1 - 0.067 = 0.993\text{ (99.3\%)}$ At 30×: $P = 1 - e^{-30} \approx 1.000$ (>99.999%)

Why this step? The exponential decay shows diminishing returns: going from 5× to 10× adds only 0.7% more coverage, but doubles cost. This math guides experimental design.

[!example] Illumina Sequencing-by-Synthesis (Most Common NGS)

Step-by-step with WHY:

1. Library Preparation

• What: Fragment DNA to ~300–500 bp, ligate adapters (short synthetic oligos) to both ends.
• Why fragment? Polymerase can only extend ~500 bp with high fidelity in clustered conditions.
• Why adapters? Provide binding sites for primers (step 3) and enable attachment to flow cell.

2. Cluster Generation (Bridge Amplification)

• What: Attach fragments to flow cell surface coated with oligos complementary to adapters. Each fragment bends ("bridges") to adjacent oligo, primed and extended → doubling the copies each cycle. After only ~10–12 cycles the cluster plateaus at ~1000 identical copies (physical crowding on the ~1 µm patch limits further doubling; you do NOT run 30 cycles or you would get ~$2^{30} \approx 10^9$ molecules, which the tiny cluster area cannot hold).
• Why amplify? A single molecule's fluorescence is too dim to detect. 1000 copies boost signal 1000×.
• Why bridge? Physical separation maintains spatial resolution—each cluster is ~1 µm apart, resolved by camera.

3. Sequencing Cycles

• What: Add fluorescently labeled reversible terminator nucleotides (A/T/G/C each with unique dye + blocking group). Flow across all clusters → incorporates one base per cluster. Image entire flow cell → record color at each position. Cleave terminator + dye → repeat.
• Why terminator? Ensures only one base added per cycle (otherwise polymerase would race ahead unevenly).
• Why reversible? Must restore3'-OH group for next cycle.
• Example: Cycle 1: Cluster A glows green (C) → Cycle 2: red (T) → Cycle 3: green (C) → read = CTC...

4. Base Calling

• What: Software converts intensity images (4 channels: A=green, C=blue, G=yellow, T=red) to FASTQ files with quality scores.
• Why quality scores? Each base call has error probability (~0.1%–1%); Phred scores ($Q = -10 \log_{10} P_{\text{error}}$) inform downstream analysis (e.g., filter bases with $Q < 30$).

5. Alignment/Assembly

• What: Map reads to reference genome (BWA-MEM) or assemble de novo (SPAdes).
• Why reference? For resequencing (variant calling), leverages known structure. For novel genomes, must assemble overlaps.

[!example] Ion Torrent (Semiconductor Sequencing)

Difference from Illumina: Detects pH change, not fluorescence.

How: Nucleotide incorporation releases H⁺: $\text{DNA-3'-OH} + \text{dNTP} \xrightarrow{\text{polymerase}} \text{DNA-3'-dNMP} + \text{PPi} + \text{H}^+$

• What: Flow one dNTP type at a time (e.g., only dATP). Base pairing rule: A pairs with T. So wherever the template has a T, polymerase adds an A → releases H⁺ → drops pH. Sensor detects voltage shift proportional to #H⁺.
• Why this works? No labels needed → faster, cheaper reagents. But homopolymers (AAAAA) are hard: 5 A's release 5× H⁺, but signal isn't perfectly linear.

Example: Template read 3'→5' = TTACG (the strand being copied). Complementary bases added:

• Flow dATP: template positions with T → 2 A's incorporated at the leading TT → large signal (2 H⁺).
• Flow dTTP: template A → 1 T incorporated → small signal.
• Flow dGTP: template C → 1 G incorporated → small signal.
• Flow dCTP: template G → 1 C incorporated → small signal.

Why this step? The signal appears when the flowed nucleotide matches (base-pairs with) the template base. dATP produces signal precisely where the template is T, because A–T pair.

[!mistake] Common Misconceptions

Mistake 1: "NGS reads are longer than Sanger"

Why it feels right: "Next-gen" sounds more advanced, and Sanger reads are ~800 bp. Why it's wrong: Illumina reads are 50–300 bp (shorter!) because paralelization trades length for throughput. The "generation" refers to throughput, not read length. Fix: Use long-read tech (PacBio, Nanopore: 10–100 kb) for structural variants or assembly. Use Illumina for SNPs/indels where short reads suffice.

Mistake 2: "30× coverage means every base is read30 times"

Why it feels right: "30×" literally sounds like 30 reads per position. Why it's wrong: Coverage is an average. Random sampling creates variance—some regions get 50×, others get 10×. Under a Poisson model with mean $\lambda = 30$, only about 1–2% of positions fall below 20× (and only ~2–3% below 22×). The point stands—coverage is uneven—but the shortfall is smaller than people intuit. Fix: Use uneven coverage plots to identify gaps; require minimum depth (e.g., ≥10×) for variant calls.

Mistake 3: "More coverage always improves accuracy"

Why it feels right: More data = better. Why it's wrong: Sequencing errors are systematic in some contexts (PCR duplicates, GC bias). At 100×, you're resampling the same errors, not correcting them. Fix: Pair depth with orthogonal QC (e.g., long reads to phase SNPs, optical mapping for structural variants). Diminishing returns start ~50× for most applications.

[!recall]- Explain to a 12-Year-Old

Imagine you have a 1000-page book, but someone ripped out all the pages and shredded them into confetti. Your job: figure out what the book said.

Old way (Sanger): Pick up one piece of confetti, read it carefully with a microscope, write it down. Pick up the next piece. This takes forever!

NGS way: Spread out millions of confetti pieces on a giant table. Take a photo where every piece glows a different color depending on its first letter. Now you know the first letter of all pieces at once. Change the colors to show the second letter → take another photo. Repeat 150 times → you've read 150 letters from each piece simultaneously.

Then use a computer to find overlaps: "This piece ends with '...the cat' and that piece starts with 'cat sat..' → they connect!" Do this millions of times → reconstruct the book.

Why is this faster? Instead of reading one piece for10 seconds each (10M pieces = 3 years!), you read them all in parallel (150 photos = 1day).

[!mnemonic] NGS = "Naive Granny Sequences"

• Naive: Fragments DNA randomly (doesn't pick specific regions)
• Granny: Goes in parallel (like knitting 1000 scarves at once)
• Sequences: By synthesis (builds DNA while reading)

Or: "Lots of Clusters In Parallel" → Library, Cluster, Image, Process

Connections

• Sanger Sequencing - first-generation method; NGS paralelizes it
• DNA Polymerase - enzyme used in sequencing-by-synthesis
• Bioinformatics Alignment Algorithms - BWA, Bowtie map NGS reads
• Genomic Variants (SNPs, Indels) - NGS identifies these at population scale
• CRISPR - NGS verifies edits and off-target effects
• Metagenomics - NGS enables sequencing mixed microbial communities
• Cancer Genomics - tumor-normal NGS finds somatic mutations
• Personalized Medicine - NGS makes whole-genome sequencing affordable

#flashcards/biology

What does NGS stand for and what is its key advantage over Sanger sequencing? :: Next-Generation Sequencing; it sequences millions of DNA fragments in parallel simultaneously, versus Sanger's one-at-a-time approach, reducing cost from ~$100M to <$1000 per genome.

Concept Map

Hinglish (regional understanding)