Paired Reads

Paired reads

Paired-end sequencing reads each DNA fragment from both ends, giving two reads per fragment: mate 1 and mate 2, usually stored in two files, R1 and R2. Everything downstream assumes the files stay in step, record for record. z-fastq treats a pair as one unit everywhere: it never keeps, drops, or moves one mate without the other.

A pair is one record from each of two files, or two neighbouring records in one interleaved file (Interleaving).

Check pairs

z-fastq check --paired R1.fastq.gz R2.fastq.gz
z-fastq check --interleaved interleaved.fastq.gz

Each mate is first validated as a record (Validation), then the two names are compared. In an interleaved file the first two records form the first pair, the next two the second, and so on.

• P001: the mates' names do not match.
• P002: a mate is missing: one file has more records than the other, or an interleaved file ends on an odd record.

Structural record errors are detected while reading. Missing-mate detection precedes semantic validation, so an orphan record can report P002 even if its bases are invalid. When both mates are present, record validation precedes name matching. Messages give the pair number, counted from 0, and shorten very long names.

How names are matched

Sequencers mark the mate in the read name in one of two ways: a /1 or /2 at the end of the name, or a second word such as 1:N:0:1 (Illumina's Casava format). The default policy, --pair-names illumina, understands both:

R1 header	R2 header	Pair?
@read42/1	@read42/2	yes
@read42 1:N:0:1	@read42 2:N:0:1	yes
@read42	@read42	yes
@read42/2	@read42/1	no, mates reversed
@read42/1	@read42	no, marker on one side only
@read42/3	@read42/4	no, /3 is part of the name

Step by step:

1. Take the first word of each header, and drop a final /1 or /2 from it.
2. Read a mate marker from the second word, if there is one: a leading 1: or 2:, or a final /1 or /2.
3. The identifiers left must be equal and not empty.
4. If both headers carry markers, mate 1 must say 1 and mate 2 must say 2, and markers within one header must agree. A marker on one side only is a mismatch.

Anything after the marker word is ignored. So @read42 1:N:0/3 and @read42 2:N:0/4 do pair: their first words agree, and their second words say 1 and 2.

--pair-names exact is the strict alternative: the whole first word must match byte for byte, and nothing is removed.

Related pages

• Sampling: subsets that keep both mates or neither.
• Interleaving: between two files and one interleaved file.
• Processing paired reads: every pair is validated before it is trimmed.
• Troubleshooting: when names do not match.