Cell Barcodes, UMIs, and Gene Identifiers

Three Labels, Three Different Questions

A sequencing read from a single-cell experiment may carry information about a cell, a captured RNA molecule, and the genomic feature from which that molecule originated. These are different kinds of information. A cell barcode answers “which cell or cell partition is associated with this read?” A unique molecular identifier (UMI) answers “which captured molecule, within that cell, does this read represent?” A gene identifier answers “which gene or other annotated feature is associated with the sequence?” Keeping these roles separate is essential: the three labels are not interchangeable, and they do not necessarily come from the same part of the data.

In droplet-based protocols, a barcode sequence is commonly attached during library preparation to identify a droplet or its associated cell, and a UMI may be attached to a captured molecule. In plate-based protocols, a well label can identify the cell’s location; the library may or may not include a cell barcode or UMI, depending on the assay. As covered in the introductory material on droplet encapsulation and plate-based capture, the experimental design determines which identifiers are present and how they are recorded.

The exact read layout is protocol-specific. A sequencer may produce separate reads or indexes for different pieces of information, and the order and length of barcode and UMI sequences vary. Therefore, a string of bases cannot be identified as “the cell barcode” or “the UMI” from its appearance alone. Its position and meaning must be interpreted using the protocol’s documented design. A sample index is another distinct label: it identifies a library or sample in a pooled sequencing run, not necessarily an individual cell.

How the Identifiers Organize the Data

Conceptually, an annotated observation can be associated with a tuple: a cell label, a molecule label within that cell, and a feature label. The labels organize different dimensions. A cell barcode does not say which gene was expressed; a UMI does not identify the cell unless interpreted alongside the cell barcode; and a gene identifier does not, by itself, say which cell contributed the sequence.

This tuple is a useful way to reason about a record, not a guarantee that every assay supplies all three fields or that every possible combination occurs. For instance, full-length plate-based assays may associate reads with a well and a gene without using UMIs. A gene label may also refer to an annotated feature other than a protein-coding gene, depending on the reference annotation. The experimental protocol and annotation define what the labels mean.

Cell Barcodes: Identifying a Cell-Associated Partition

A cell barcode is designed to distinguish cell-associated partitions in a library. In droplet protocols, barcode sequences are often drawn from a predefined set, sometimes called a whitelist. The barcode sequence is then used to associate reads with a partition represented in that set. A barcode is not a biological property of the cell, such as cell type or donor identity; it is a technical label assigned by the assay.

A barcode does not prove that a partition contains exactly one cell. A partition may contain no cell, one cell, or more than one cell, and barcode sequence errors can also complicate interpretation. Those are separate questions from what the barcode labels. Likewise, if the same barcode sequence is used in different libraries, it need not identify the same physical cell across those libraries. Sample or library context may be required to distinguish reused labels.

For plate-based capture, a well position may provide the cell label without a cell barcode being sequenced in the read. A sample index, in turn, can identify which pooled library a read came from. These labels operate at different levels: a well identifies a position within a plate, a cell barcode identifies a cell-associated partition under the protocol, and a sample index identifies a library or sample in a sequencing pool.

UMIs: Identifying Captured Molecules

A UMI is attached to a captured molecule so that reads carrying the same cell-associated UMI can provide evidence about the same original molecule. Its purpose differs from that of a cell barcode: many distinct molecules can come from one cell, so the cell barcode alone cannot distinguish them. A UMI is interpreted in the context of the cell label and, where relevant, the associated feature. The UMI sequence is a technical tag, not a gene name and not a cell identity.

UMIs have a finite sequence space. If a UMI has length \(k\) and uses four possible bases at each position, the number of possible sequences is \(4^k\), assuming all sequences are allowed. For a five-base UMI, that gives \(4^5 = 1{,}024\) possible sequences. Different molecules can therefore receive the same UMI by chance, especially as the number of molecules grows. Sequencing errors can also make one original tag appear as a different observed sequence. Thus “unique” describes the intended role of the identifier, not a guarantee that every molecule has a globally unique, error-free tag.

Gene Identifiers: Labels Inferred From a Reference

A gene identifier names a feature in a reference annotation. Depending on the chosen annotation, it may be a stable database-style identifier, a gene symbol, or another feature label. Symbols can be convenient to read, while stable identifiers can reduce ambiguity when names change or are shared. The identifier’s interpretation depends on the reference annotation used; the same sequence may be treated differently under different annotations, especially near feature boundaries or for overlapping features.

In most workflows, the RNA-derived sequence is compared with a reference genome or transcript set, and the associated gene label is determined from that comparison and annotation. The gene identifier is therefore not usually a third sequence tag physically attached to the molecule. It is a computational label assigned from sequence evidence. The detailed process of read alignment and gene assignment is a separate step; here the key distinction is between observed barcode sequences and an inferred feature label.

Checks, Alternatives, and Common Errors

A practical check is to ask three separate questions: what physical or computational item does this label identify, where did the label come from, and what context is needed to interpret it? Protocol documentation should establish the read layout, barcode design, and whether UMIs are present. Reference and annotation documentation should establish how gene labels are defined.

The main alternatives are not interchangeable versions of one identifier. Droplet assays commonly use cell barcodes, whereas a plate-based assay may rely on a well identifier. UMI-based assays carry molecule labels; non-UMI assays do not, so their read records do not contain a UMI field to interpret. Gene symbols and stable gene identifiers are alternative naming systems for annotated features, but neither replaces the cell or molecule label.

Warning signs include interpreting a sample index as a cell barcode, treating a gene symbol as a sequence read, assuming every barcode represents one cell, or assuming UMIs cannot repeat. A barcode sequence that is unexpected or differs slightly from a documented barcode may reflect a sequencing error, but the sequence alone does not establish the correct interpretation. The protocol and its processing rules are needed.

This distinction matters because later summaries depend on knowing which dimension a label represents. Confusing the labels can merge different cells, treat repeated sequencing of a molecule as distinct molecules, or assign biological meaning to a technical sequence. The safest practice is to preserve each field separately and record the protocol and reference context that give it meaning.