A glossary of sequencing terms

People talk past each other in sequencing meetings because coverage and depth are not the same noun, and because a VCF line is not a read. This glossary gives working definitions for the words that show up between a library and a claim. Each definition includes the mistake it is meant to prevent. The physical path from library to read is next-generation sequencing from library to reads. How those files become a result you can argue with is from reads to a result a biologist can question. The definitions below are research vocabulary. They do not certify an assay.

A read is one observation, and a contig is a consensus stretch

A read is the instrument's record of one pass along a fragment: a string of bases plus, for most short-read and many long-read formats, a quality for each base. One read is not one independent molecule of the genome if PCR made many copies and the sequencer saw several of them. Treating every read as an independent biological event is the mistake this word prevents. When you mean the fragment that went into the library, say fragment or molecule, and then show how you collapsed copies.

A contig is a contiguous sequence built by assembly, a consensus of overlapping reads. It is not a chromosome, and it is not a proof that the genome is finished. The mistake it prevents is quoting a contig length as the size of a chromosome arm, or treating a gap between contigs as a biological deletion. Gaps are often places the assembler refused to guess. Sanger sequencing still produces a kind of read, a trace of one reaction. Calling that trace a contig confuses a single-molecule chromatogram with an assembly.

Coverage, depth, duplicates, and insert size

Coverage, used carefully, is breadth: the fraction of the reference that received reads. Depth is the count at a base, and mean depth is that count averaged over a region you must specify. The mistake "30x coverage" prevents, when you unpack it, is believing the whole genome is uniformly 30 reads deep. It may be 30 on average and zero in the repeats. Say mean depth and say breadth when you need both.

A duplicate is a read pair, or a read, judged to be a copy of another from PCR or from an optical accident on the flow cell. The mistake it prevents is counting those copies as extra evidence. Marking duplicates and then still using them to call a rare allele invents support. Some applications, such as certain counting assays, have their own rules. Write the rule down. Do not inherit a variant-calling habit.

Insert size is the length of the genomic fragment between the adapters, not the length of the read. In paired-end sequencing the two reads spy on the ends and the insert size is inferred. The mistake it prevents is designing a "150 base pair insert" when you meant a 150 base pair read, and then wondering why mates overlap or why they span a different distance from the one the structural caller expected.

Phred, MAPQ, adapter, and index

A Phred score is a base quality. In the common definition, a score of Q means the estimated probability of an error at that base is ten to the power of minus Q over ten. A high score is confidence in the base call. It is not confidence in where the read belongs. The mistake it prevents is throwing away a whole read because one base is weak, or, in the other direction, trusting a variant base solely because the quality character looked impressive in a text file.

MAPQ is a mapping quality, a score for how confidently the aligner placed the read. Low MAPQ is what repeats produce when several genomic homes look equally good. The mistake it prevents is summing a multi-mapping read into a depth track as if it had one true address. Filters that ignore MAPQ turn repeats into false peaks and false variants.

An adapter is the synthetic sequence added so the fragment can bind the instrument's chemistry. It is not biological sequence. The mistake it prevents is aligning adapter bases to the genome and calling them a variant at the end of a read. Trim adapters, and record that you did. An index, sometimes called a barcode in the sample-sheet sense, is a short sequence that marks which sample a read came from. It is not an adapter, though it is often read from the same synthetic end, and it is not a unique molecular identifier unless your protocol added one on purpose. The mistake it prevents is demultiplexing on the wrong bases and then analysing a mixture under a single name. Collisions between indexes are treated in index collisions in multiplexed pools.

VCF, BAM, CRAM, and demultiplex

A VCF is a variant call format file: positions where the sample is claimed to differ from a reference, with fields for genotype, quality and filters. The mistake it prevents is treating the file as raw data. It is an argument. A different caller, or a looser filter, produces a different VCF from the same reads. Coordinates assume a reference build.

A BAM is a binary alignment of reads to a reference. A CRAM is a more compressed alignment that typically needs that same reference to be expanded. The mistake both terms prevent is archiving "the alignment" without the build, and the further mistake of assuming a CRAM can be opened years later on a machine that no longer has that genome. A FASTQ is the usual file of unaligned reads and qualities. Demultiplex is the split of a pooled run into per-sample files using the index. The mistake it prevents is believing the split checks biology. It checks the index against the sheet. A wrong sheet demultiplexes cleanly into the wrong names.

You can view a coordinate from these files on the UCSC Genome Browser only after the build matches. Reads you later share may belong in the Sequence Read Archive, under the consent you have, which is a policy decision and not a property of the file format.

Where the words sit on a pile of reads

Picture one locus. Several reads overlap a base. Some are marked duplicate. One aligns with a low MAPQ because the next exon looks similar. Adapter sequence, if you forgot to trim, dangles off the end and fails to match. The depth you quote depends on which of those reads you keep. The glossary is a way to say that sentence in labels.

Using the words when a report goes wrong

If a collaborator says the sample has excellent quality, ask for the Phred distribution and the MAPQ distribution separately. If they say the gene is covered, ask for breadth over that gene and for depth at the exon you care about. If they say the variant is in the VCF, ask which filter it passed and whether the BAM still shows it on both strands. If a demultiplex report looks perfect and the biology looks swapped, the glossary will not mention the plate map, but sample identity mix-ups and plate maps will. The words stop you curing an identity problem with a stricter Phred cut.

The short-read method class these formats grew up around is outlined by Illumina. Long reads use the same nouns with different error shapes. Do not copy a short-read MAPQ habit onto a long-read alignment without reading what that aligner means by the score.

Research vocabulary is not a clinical grade

None of these terms turn a file into a diagnosis, a biosafety clearance, or an approved test. A VCF of a human sample in a research project remains a research file. Reporting it to a patient is a different activity, under different rules, and this glossary does not authorise it. Handle the biological sample under your institution's rules before any of the files exist.

Shared drives and mixed dialects

When a core facility and a research group in another city exchange files, they often use coverage for depth and barcode for index. Agree the glossary in the statement of work before the drive is copied. A partial download after a dropped connection should be checked against a checksum, and the file names should already contain the build and the filter. A folder called final_final is how the wrong VCF becomes the archive. The statement of work is the place those nouns get pinned down.

How to use the words in an enquiry

Say the deliverable with the glossary's nouns: FASTQ, BAM or CRAM against a named build, VCF with the filter described, and demultiplex rules for the indexes. Say whether mean depth or breadth is the acceptance metric. Consumables that feed the run are in the genomics and sequencing catalogue. The research context is the genomics research pathway. A genome-scale file set can be discussed through the whole-genome sequencing enquiry reference. A single-locus trace, which is a different kind of read, can be discussed through the Sanger DNA sequencing enquiry reference. Put the vocabulary in the quote request so the files that come back still answer the question you asked.