Module 1: Sequence Data & Quality Control

The first rule of sequence analysis: look at your reads before you trust them.

Learning objectives

  1. Read a FASTQ record and interpret its quality line.
  2. Convert between Phred scores and error probabilities.
  3. Identify adapter contamination and decide what to trim.
  4. Explain why aggressive trimming can be worse than none.

The FASTQ record

Four lines per read:

@hbv_TR_R02B35_read1          <- identifier
GCTGCTATGCCTCATCTTCTTGTTGGT   <- sequence
+                             <- separator
IIIIIIIIIIIHHHHHGGGGFFFEEDD   <- quality, one char per base

Line 4 is the one people ignore. Each character encodes a Phred score:

\[Q = -10 \log_{10} P(\text{base is wrong})\]

Q Error probability Meaning
10 1 in 10 Poor
20 1 in 100 Acceptable
30 1 in 1,000 Good
40 1 in 10,000 Excellent
NoteThe scale is logarithmic

Q30 → Q20 is a tenfold increase in error rate, not a 33% one. Small-looking drops in mean quality matter more than they appear.

What to look for

  • Quality by position. Quality falls toward the 3’ end on Illumina — that is normal chemistry, not a failed run.
  • Adapter contamination. When the insert is shorter than the read, you sequence into the adapter. These bases are real sequence, but not your sequence, and they will not align.
  • Duplication. PCR duplicates inflate apparent depth without adding information. Ten reads from one original molecule is not tenfold evidence.
  • Length distribution. For long reads, a large spread is expected.

The trimming trade-off

WarningAggressive trimming causes its own problems

Trimming at Q30 discards real data and biases you toward the reference: reads carrying true variants often have slightly lower quality at the variant position, so hard filtering can systematically remove the variants you are looking for.

Trim adapters always. Trim quality gently. Q20 is usually plenty.

Long reads are different

Nanopore data has higher per-base error, and it is not uniformly distributed — homopolymer runs are systematically harder. Five A’s may be called as four. This is a known failure mode and no amount of coverage fixes a systematic error.

The compensation is read length: a 3kb read spanning the whole HBV genome resolves structure that short reads cannot.

Exercises

  1. Take a FASTQ from the workshop data. What is the mean quality? Where does it drop?
  2. Convert Q25 to an error probability by hand.
  3. Trim at Q20 and Q30. How many reads survive each? What did Q30 cost you?
  4. Discussion. Why might a variant caller miss a real variant in a homopolymer?

Takeaways

  • Look at quality before analysis, every time.
  • Trim adapters always; trim quality gently.
  • Nanopore errors are systematic, not random.