Next-generation sequencing from library to reads

Next-generation sequencing reads many DNA fragments at once and writes a base call for each cycle it can see. The library is fragments of a chosen size, fitted with adapters the instrument can hold and prime. A run is only as relevant as that library. This page follows the path from a nucleic-acid tube to a read file. It is a research explainer, not a manufacturer's run manual and not a clinical sequencing protocol.

The instrument class many benches picture is short-read sequencing by synthesis, with fluorescent reversible terminators and a flow cell, as described in the public Illumina technology overview. Long-read platforms ask different questions and use different libraries. Do not mix their quality rules into a short-read figure. Catalogue classes for the work sit under genomics and sequencing. The requirement travels with the quote request.

What a library is

Genomic DNA, or amplicons, or cDNA, is not yet a library. Library preparation breaks the material into fragments the chemistry can handle, repairs the ends so adapters can ligate, and attaches those adapters. The adapters carry the sequences the flow cell binds and the primer sites the sequencing reaction uses. In a multiplexed run they also carry indexes, so reads from different samples can be sorted after the fact.

Fragmentation is mechanical or enzymatic. The size of the insert decides whether paired reads will overlap, whether a repetitive region is spanned, and whether clusters amplify well. A library of very short inserts answers a different question from a library of long ones.

PCR during library preparation increases material and also copies whatever amplified most easily. Those copies become duplicate reads. PCR-free libraries avoid that step when the input is large enough. Unique molecular identifiers help only when the analysis actually uses them to tell copies from starting molecules.

Adapter dimers are short adapter-adapter molecules with almost no insert. They cluster very efficiently and produce reads that are mostly adapter. A size trace that shows a sharp peak near the length of two adapters is a warning to clean the library again, not a minor cosmetic. Primer or adapter leftovers are the same family of problem as primer-dimer on a PCR gel.

From molarity to a flow cell

Instruments consume a number of molecules, not a mass in a notebook. A mass measurement that ignores the average fragment length will overload a library of short pieces and underload a library of long ones. Fluorometric assays that see double-stranded DNA, plus a size distribution, are the usual pair. Absorbance alone cannot see adapter dimer and cannot tell DNA from free nucleotides. The habit of trusting one number is how flow cells are wasted.

Too many molecules crowd the clusters so their signals overlap and base calling fails. Too few waste the flow cell. The acceptable window belongs to the instrument and the chemistry version you are running. A number copied from a different model is a common route to a dim run.

Indexes need a plan. If two samples share an index, they cannot be separated. Index hopping, where a barcode is misassigned to another sample's fragment, is a known multiplex failure. Unique dual indexes, when the chemistry supports them, make that hop visible and filterable. A water-only library, or a blank carried from extraction, shows you the contamination and hopping floor of the run. Extraction blanks are discussed with the logic in how DNA extraction methods differ.

A control library or a known spike-in separates an instrument failure from an empty sample library. A failed control means the sample reads should not be analysed as if the cycle images were fine.

What the read file does and does not contain

Each cluster produces a read, or a pair of reads, with a quality score per base. Those scores are the caller's confidence. They drop for the usual chemical reasons: a difficult sequence context, a fading cluster, a cycle that was disturbed. FastQC is a public tool that plots per-base quality, adapter content, duplication and a few other signs that the FASTQ is not what you hoped. It does not trim, map, or decide whether a variant is real. A passing plot is permission to start analysis, not the end of it.

Mapping places reads on a reference. The reference has a version. Ensembl and GenBank are public homes for those sequences. A variant called against the wrong assembly is a bookkeeping error that looks like biology. Unmapped reads can be contamination, circular DNA the reference omitted, or a library so full of adapter that there was nothing to map. Look at them before you discard the fraction as "low quality DNA" in the sample.

Coverage depth and coverage breadth are different. Reads piled on one exon can be an amplicon or a capture failure. A genome-wide claim needs breadth, and a variant claim needs depth at that locus plus an error model. PCR duplicates make depth look higher than the number of independent molecules.

Choosing the shape of the experiment

Whole-genome libraries, targeted capture and amplicon sequencing spend reads on different fractions of the sample and fail for different reasons. A single sequencing label does not say which claim you ran. Amplicon runs inherit PCR bias and any contaminated no-template well.

Sanger sequencing remains the straightforward check for one plasmid or one edited amplicon. The Sanger DNA sequencing enquiry reference is a way to specify that job. A whole-genome question is a different specification, and the whole-genome sequencing enquiry reference is a prompt for it. Both are independent method references. Ask whether a quotation is possible. Neither page means a sequencer is already running for you.

A swapped index spreadsheet is indistinguishable from a surprising biological result. Demultiplex from the sheet that was written when the tubes were labelled.

Rooms, power and the limit of the claim

A library left on a warm bench is being incubated. Keep the instrument room inside the range stated for that instrument, and do not invent a new loading concentration because the room is hot. Let a cold flow cell meet room temperature with the lid closed before you open it. Condensation is water you did not mean to add.

A power cut during a run is an instrument event, not a minor pause. Some platforms can resume in ways their own manual describes. A resumed run is not automatically comparable to an uninterrupted one. Record it, and do not pool those reads with a later batch as if the cycle history were the same. Data files need a copy that survives the same power cut. A FASTQ that exists only on the instrument workstation is one disk away from disappearing.

Research sequencing of human or animal samples stays inside the ethics approval and the biosafety level of those samples. Reads are not a diagnosis. They are not a licence to identify a person outside the consent that covered the tube. This article does not provide that consent framework.

What the enquiry has to specify

EVRINTH can take a sourcing question. State the organism, the library type, the read length and pairing you need, the number of samples, the input integrity, and whether the nucleic acid is already a library or still a lysate. The genomics research pathway connects that specification to the surrounding design. Ask whether a quotation is possible. Include the reference genome you intend to map against. A request that says only "sequence this" has not yet chosen a method.