Sequencing reads

In one sentence

A sequencing read is an instrument-derived sequence of bases from a prepared molecule, with quality information that helps assess how reliably those bases were identified.

The intuition

Imagine reconstructing a book from photographed snippets. A read is one snippet; alignment proposes where it belongs. A clear photograph can still fit several nearly identical passages. The analogy stops at the chemistry: preparation can copy, shorten or selectively recover molecules before the instrument ever sees them.

How it works

A sequencing library is prepared material derived from a specimen. In a common short-read DNA workflow, DNA is fragmented, adapters are added, and fragments are sequenced. Many RNA workflows first make DNA copies of selected RNA molecules. Long-read methods recover longer sequences using different preparation and measurement strategies.

Base calling translates instrument signals into letters. A FASTQ file commonly stores a read identifier, sequence and per-base quality values. A Phred quality score is a logarithmic estimate of base-call error: Q20 corresponds to an estimated 1% error probability for that base, and Q30 to 0.1%. These are model-based estimates, not probabilities that a reported mutation is biologically real. Ewing and Green established the error-probability framework.

Alignment compares a read with a reference genome or transcript collection. Mapping confidence concerns the proposed location; base quality concerns a letter. Repeated regions can make placement ambiguous even when the letters are clear.

In paired-end sequencing, two reads come from opposite ends of one library fragment. They provide complementary placement information, but are not two independent original molecules. Amplification can also make multiple copies of a fragment. Recognizing duplicates or using validated molecular tags helps assess independent support; neither automatically repairs an error introduced before tagging.

Why it matters in cancer

A small tumor population may contribute few variant-containing molecules. Many reads copied from those few molecules cannot replace genuinely independent observations. Meanwhile, a sequence change or fusion can be difficult to align. Before trusting a finding, ask what the reads support, where they map, and how many original fragments they represent.

Assay card

Field	What to retain
Measures and method	Sequence observations produced by library preparation, base calling and alignment
Input and tissue cost	DNA or RNA extracted from a specified specimen; extraction and library preparation consume an aliquot, while reanalyzing existing files consumes no new tissue
Output and units	Sequences, read lengths in bases, quality scores, read or read-pair counts, and alignment information
Thresholds	Base, mapping and fragment-support filters set for the particular platform, pipeline and intended question; Q30 is a quality scale, not a universal diagnostic pass rule
Failure modes	Contamination, damaged input, amplification artifacts, adapters or ambiguous placement
Limits	A read does not identify a cell, establish function, or by itself prove a somatic mutation
Validation context	Check the specimen types, change classes and full workflow validated for the reported use

Common confusions

Read versus molecule: paired ends and amplification copies can share one original fragment.
Base quality versus mapping quality: a confidently read letter can occupy an uncertain location.
Raw versus usable reads: filtering changes the denominator used for downstream support.

Try it

A fictional finding has ten supporting reads, comprising five overlapping read pairs from five library fragments. Are there ten independent fragments?

Answer: No. There are at most five represented library fragments. Duplicates may reduce that number further. Ask how the workflow establishes independent molecular support.

Depth and coverage: how observations are distributed.
Variant calling: how sequence evidence becomes a candidate finding.

Sources

Source check: October 9, 2026. Expert and learner review remain pending. Examples are fictional.