From Sequencing Libraries to Short Reads

From DNA Molecules to Recorded Reads

A sequencing file is not a direct photograph of a genome. It contains records produced by a chain of laboratory and computational steps: genomic DNA is broken into fragments, prepared so that a sequencer can process it, read in cycles, and converted by a base caller into strings of letters with quality scores. Each step affects what can eventually be observed. To interpret a short-read file, it helps to keep three objects distinct: the original DNA molecule, the prepared library molecule, and the read recorded by the instrument.

The outline below describes a common short-read workflow, not a universal recipe. Library preparation and sequencing chemistry differ across platforms and protocols. Details such as end repair, amplification, and how a molecule is attached for reading are method-dependent; the general distinction between prepared molecules and base-called reads remains useful across them.

Preparing the Library

In many short-read workflows, DNA is fragmented into pieces that fall within a useful size range. Fragmentation may be physical or enzyme-based. The resulting pieces can have ends that need preparation before adapters can be attached. Depending on the protocol, end repair or other processing may occur, and adapters are then ligated to the fragment ends.

Adapters are known sequences designed to let library molecules interact with the sequencing system. They provide binding or priming sites needed for later steps. Some protocols also attach an index (also called a barcode) to help identify which sample a library came from when multiple samples are processed together. An index is not part of the sample’s genomic sequence. It is a deliberately added label, and its exact location and use depend on the protocol.

Some workflows amplify library molecules before sequencing. Amplification can provide more material, but it can also change the representation of the original fragments: some molecules may be copied more readily than others, and copying may introduce errors. These possibilities arise before a read is produced. They are different from a base-calling error, which is an incorrect letter assigned to a signal during sequencing.

Prepared molecules are made available to the instrument. In a common short-read approach, a molecule is copied or otherwise presented in a form that produces measurable signals over successive sequencing cycles. The instrument records signals; a base caller converts those signals into a sequence of letters and estimates a quality for each called base. In this way, the final record depends both on the DNA in the sample and on preparation, measurement, and computational interpretation.

What a FASTQ Record Contains

A common raw-read file format is FASTQ. Each record has four lines: a line beginning with an identifier, a line containing the called sequence, a separator line, and a line of quality characters. The quality line has one encoded value for each sequence position. The identifier may contain information about the read, instrument run, or sample index, but the precise fields vary. Paired-end data—where both ends of a library molecule are read—are commonly stored as corresponding records in two files or in an interleaved file.

The sequence line records the base calls, usually represented with A, C, G, and T, and sometimes with a character such as N when the base caller does not assign a definite base. The quality line does not give a second sequence. It encodes the base caller’s estimated confidence in each position. The scores use the Phred scale: if \(e\) is the estimated probability that a base call is wrong, then \(Q=-10\log_{10}(e)\), or equivalently \(e=10^{-Q/10}\). These are estimates from the base-calling process, not guarantees about any one base.

A paired-end record represents two reads from opposite ends of the same prepared library molecule. That does not mean the two reads together always reveal every base in the original fragment. If the insert is longer than the combined read lengths, an internal part of the fragment is not directly read; if it is shorter, the reads may overlap. The fragment itself may also include adapter sequence in the recorded read when the sequenced insert ends before a read cycle does. These are consequences of the relationship between library construction and read length, not evidence that a read file contains a complete copy of each source molecule.

Where Technical Problems Can Enter

Errors and distortions can arise at different stages, and the distinction matters when interpreting a read file. Fragmentation may produce a nonuniform set of fragment lengths. Ligation may be inefficient for some molecules, while adapter dimers—adapter sequences joined without a genomic insert—can enter the prepared library. Amplification can favor some library molecules over others or introduce copying errors. During sequencing, signal measurement and base calling can produce incorrect or uncertain calls. These possibilities are not interchangeable: for example, an adapter dimer is a library-preparation artifact, while a low-quality base call is uncertainty about a recorded letter.

Some practical checks follow directly from the record structure. Sequence and quality strings in a valid FASTQ record should have equal lengths, because there must be one quality value per called base. A large amount of adapter sequence in reads can indicate that the read extends beyond a short insert or that adapter removal has not yet been performed. A high proportion of uncertain calls or persistently low per-base scores can flag a sequencing or library-quality concern. These checks can reveal warning signs, but a file alone may not identify the cause; the protocol, run information, and laboratory records provide important context.

How to Interpret the File—and Its Limits

A FASTQ file answers a focused question: what sequence did the base caller assign to each read, and what quality score did it assign at each position? It does not, on its own, say exactly which original genomic molecule produced every record, whether every molecule in the sample was equally likely to enter the library, or whether a called difference from a reference is biologically real. Those questions require additional processing and evidence. For now, the important point is that sequence records are observations generated by a workflow, not unmediated copies of the sample’s DNA.

Short-read sequencing also differs from a genotyping array. An array measures signals at selected sites determined by its probe design; a sequencing read records a short stretch of called bases from a prepared molecule. Neither measurement is free of technical limitations, but their measurement mechanisms and resulting data are different. A read should not be treated as an array genotype call, and an array’s selected probe set should not be mistaken for a set of sequence reads.

Paired-end sequencing is one option when information from both ends of a library molecule is useful; single-end sequencing records only one end. Paired-end reads can provide more observations of a fragment’s ends, but they do not automatically reveal the unsequenced middle of a long insert. Similarly, increasing read length may change how much of an individual fragment is observed, but it does not eliminate biases introduced during library construction or uncertainty in base calling.