Lecture 1.2

The digital genome

When a genome is sequenced, it does not come as a simple string of letters. Instead, you get millions of short reads, and each base has a chance of being incorrect. To show this uncertainty, we use the FASTQ format and Phred quality scores. These help us decide which data to keep. These scores use a log scale. For example, going from Q20 to Q30 means the error rate drops by a factor of 10, which is a big improvement. Any errors you allow at this stage will affect all later analyses. That is why filtering is the first step.

Learning objectives

After this lecture, you should be able to do the following.

Assemble a contig from overlapping reads, choose between reference-based and de novo assembly, and read what the assembled output reports.

  1. Interpret a contig FASTA header (number, length, and coverage).
  2. Interpret contig length as a signal of assembly quality.
  3. Classify a variant as a SNP or an indel.
  4. Assemble a contig from a small set of overlapping reads.
  5. Choose reference-based or de novo assembly for a given organism and goal.

Explain what a quality score records about a base call, and determine how a change in the score changes the error rate.

  1. Explain why each base call is a best guess with a chance of being wrong.
  2. Using \(P = 10^{-Q/10}\), determine how the probability that a base call is wrong changes for a stated change in Phred score.

Read the text formats that carry sequences and reads, and say which one a given task needs.

  1. Identify the four lines of a FASTQ record.
  2. Describe the FASTA format.
  3. Contrast FASTA and FASTQ, and say when each is used.

Read a FastQC report, diagnose what produced a quality problem, and judge whether an unfamiliar signature is a problem at all.

  1. Interpret the Per Base Sequence Content and Per Sequence GC Content plots.
  2. Interpret a FastQC Per Base Sequence Quality plot.
  3. Explain what FastQC reports and why reads are checked before analysis.
  4. Predict whether an adapter appears in a read from the read length and insert size.
  5. Diagnose the likely cause of a quality problem from a set of FastQC plots.
  6. Evaluate whether an atypical FastQC signature reflects a real problem or a genuine feature of an unfamiliar sample.

Account for how a sequencing platform turns a DNA sample into reads, and predict what its read length costs you downstream.

  1. Explain how Illumina uses reversible terminators to read one base per cycle.
  2. Classify a sequencing method as optical, electrical, or chemical from how it reads bases.
  3. Predict how read length affects the assembly of a repetitive region.
  4. Order the wet-lab steps that turn cells into sequencing-ready DNA.

Readings

Supplemental

HDI ch. 1, §1.2
Ismail 1.2 runs through the history, Maxam-Gilbert and Sanger, then the jump to next-generation short reads and third-generation long reads. It also lays out library prep: fragmenting the DNA, repairing the ends, and ligating adapters. The thing to take away is why NGS is fast. It reads millions of fragments at once, where Sanger handled one template at a time. Notice where adapters get added; that comes back when we talk about trimming. The Maxam-Gilbert cleavage chemistry is here for context. Read it once and move on.
(T 17, 18, 19, 20)
HDI ch. 1, §1.3.2
A short section that sets up the Phred score. It gives \(Q = -10 \log_{10}(P)\) and a table for what Q20, Q30, and Q40 mean as error rates. Get comfortable with the log scale: every 10 points of Q is another factor of 10 in accuracy, so Q30 is one wrong base in a thousand. That intuition is the whole section. Sequencing depth in 1.3.1 belongs with a later coverage discussion. Skip it for now.
(T 6, 7)
HDI ch. 1, §1.4
Ismail 1.4 breaks down the four-line FASTQ record and shows how each quality score is packed into a single ASCII character with the +33 offset, with worked examples. Put the sequence line above the quality line and read them together. Being able to turn a character into its Q, and back, is the skill here. Ignore the older Phred+64 encoding; Illumina data today is Phred+33. FASTA barely appears, so use the slides for that side of {lo-fasta-vs-fastq}.
(T 8, 9, 10)
HDI ch. 1, §1.5
This is the FastQC walkthrough. It goes module by module, Per Base Sequence Quality, Per Base Sequence Content, Per Sequence GC Content, and the rest, pairing a clean report against a broken one for each. Three plots do most of the work: the per-base quality boxplots (watch the drop at the 3-prime end), per-base content (a clean library gives flat, parallel lines), and per-sequence GC (one peak at the genome average). Once you can read them, practice naming what went wrong when they look off. That last step is {lo-fastqc-diagnose}. Per Tile Sequence Quality and K-mer Content rarely decide anything. Know the names.
(T 11, 12, 13, 15)
HDI ch. 1, §1.6
Section 1.6 is preprocessing: quality trimming, length and N-content filters, and adapter removal. For this lecture the adapter part is what matters. Read it for adapters, why they end up in a read and why they have to go before you align or assemble. Treat the rest of the filtering menu as background. The exact tool flags are a lab concern, not an exam one.
(T 14)
HDI ch. 3, §3.1 and §3.3
Chapter 3 introduces de novo assembly, greedy, overlap-consensus, and de Bruijn graphs, then covers how you judge an assembly with N50 and BUSCO completeness. Keep it conceptual. Reads overlap where they share sequence, those overlaps stitch into gap-free contigs, and a bigger N50 means the pieces came together better. A pile of short contigs usually means repeats or thin coverage. The ABySS and SPAdes command runs and the graph math are more than this lecture needs.
(T 1, 2, 4, 5)
BTA ch. 4, §4.1
Brown sorts sequencing into three families, chain termination, short-read Illumina, and long-read SMRT and nanopore, and explains the chemistry behind each. The Illumina cycle is worth reading slowly: add one blocked, labeled base, take a picture, cut the block off, repeat. Then hold that against nanopore to see the length-versus-accuracy tradeoff. The capillary Sanger figures are background for a short-read course.
(T 17, 18, 19)
BTA ch. 4, §4.2-4.3
Section 4.2 covers shotgun sequencing and how reads build into contigs and then scaffolds. Section 4.3 shows the idea in action, assembling the Neanderthal genome against the human reference. Use 4.2 for the overlap-to-contig story and the contig-versus-scaffold distinction. The Neanderthal example is a clean case of when a reference makes assembly easier. The Human Genome Project history reads well but is not tested here.
(T 1, 2, 4, 5)
LT ch. 1
A platform-by-platform tour, 454, SOLiD, Illumina, Ion Torrent, PacBio, and Oxford Nanopore, with the signal, read length, and cost of each. Skim the comparison table and fix two anchors in your head: Illumina is short, cheap, and accurate; PacBio and nanopore are long but noisier. 454 and SOLiD are retired. Recognize the names and move on.
(T 17, 18, 19)
LT ch. 3
The hands-on version: run FastQC on real Illumina data, then trim and filter with fastx-toolkit, matching each bad plot to the step that fixes it. If you want the QC ideas to click, follow one dataset from its first FastQC report through trimming and watch the plots improve. Installation and Linux setup are lab logistics.
(T 11, 12, 13, 14)
GJL ch. 4, §4.3-4.4
Gonzaga-Jauregui and Lupski walk the whole workflow in 4.3, DNA prep, library prep, sequencing, then analysis, and give a generation-by-generation overview in 4.4. Read 4.3 for the order of the wet-lab steps, watching where fragmentation and adapter ligation land relative to the sequencing run itself. The whole-genome versus whole-exome cost debate is a rare-disease detour for us.
(T 14, 18, 19, 20)
GJL ch. 4, §4.5
Section 4.5 is about what you do with reads once you have them: map them to the reference and call variants, both SNPs and indels. Lock in the SNP-versus-indel distinction, a single swapped base against an insertion or deletion that shifts everything downstream in the alignment. The clinical rare-disease material further on is outside this lecture.
(T 3, 5)
BSM §4.2
This chapter puts assembly on a formal footing with de Bruijn graphs: k-mers become nodes, and the genome is a path through them. Optional. Read it if you want the algorithm under the overlap picture. For the exam, the takeaway is still that shared sequence drives assembly. The graph proofs and the annotation chapter are far past this lecture.
(T 4, 5)
HY ch. 7, §7.3
A quick, plain-language pass over genomics methods, microarrays, DNA sequencing, whole-exome, and single-cell, inside a broader high-throughput chapter. A gentle refresher if the platform overview felt too fast. It sits sequencing next to the other high-throughput methods. Transcriptomics, proteomics, and metabolomics are for later modules.
(T 18, 19)
Back to top