Spliced and Unspliced RNA Counts

What the Two Count Types Measure

Single-cell RNA sequencing can record evidence from different parts of a gene’s RNA. An alignment to an intron can indicate an RNA molecule that has not yet had that intron removed; an alignment to an exon can indicate RNA containing an exon sequence. Separating these observations into unspliced and spliced counts gives two measurements with different biological interpretations. Their joint pattern can be informative about recent transcription, but each is still an assay-derived count, not a direct reading of a transcription rate.

For a cell \(i\) and gene \(g\), write \(U_{ig}\) for the unspliced count and \(S_{ig}\) for the spliced count. These are nonnegative counts assigned by a specified alignment and quantification procedure. Depending on the assay and processing pipeline, the counted units may be reads, read pairs, or deduplicated molecules. The unit must be checked in the data; a count of 20 deduplicated molecules is not the same measurement as 20 raw reads.

A common classification uses a reference genome and gene annotation to distinguish reads aligned to introns from reads aligned to exons. An intron-only alignment is evidence for an intron-containing transcript and may be assigned to the unspliced category. An exon-only alignment may be assigned to the spliced category. A read overlapping an exon–intron boundary, or otherwise compatible with more than one category, may be labelled ambiguous and excluded from both counts or handled by a pipeline-specific rule.

This distinction is not perfect. An intronic alignment can arise from retained introns, unannotated transcripts, or genomic DNA contamination, as well as from a newly transcribed precursor. Exonic sequence can also be present in a precursor that still contains introns. Thus, “unspliced” and “spliced” are operational labels based on sequence evidence and annotation, not microscopic observations of complete RNA molecules.

Why the Pair Contains Information

The biological motivation is temporal. RNA containing introns is often a relatively recent product of transcription, while RNA with introns removed can persist after the initial transcription event. Therefore, a cell with more unspliced evidence relative to spliced evidence may have a larger recent or precursor-RNA contribution for that gene. This is a qualified interpretation: the observed pair also depends on RNA processing, degradation, capture, sequencing depth, and assignment choices.

A simple descriptive summary is the unspliced fraction among the two classified counts:

This fraction describes the composition of the assigned counts for that cell and gene. It is not the fraction of all biological RNA molecules that are unspliced unless the measurement process makes the assigned counts representative of those molecules. If both counts are zero, the fraction is undefined; it should not be reported as zero. A small denominator also makes the fraction unstable, so it is important to show or retain the underlying counts.

The connection between the fraction and the observed counts follows directly from the definition. If \(N_{ig}=U_{ig}+S_{ig}\) is the total assigned count, then \(U_{ig}=N_{ig}\widehat{f}_{ig}\) and \(S_{ig}=N_{ig}(1-\widehat{f}_{ig})\). The fraction separates the relative balance of the two categories from their total count. It does not remove differences in capture efficiency: if the assay detects intronic and exonic evidence at different rates, the measured fraction can differ from the biological RNA fraction.

At the cell level, the pair \((U_{ig},S_{ig})\) describes the evidence observed for one gene in one cell. Looking across genes can help characterize which cells show relatively more intronic evidence for particular genes. At the gene level, the pairs can be compared across cells to see whether a gene’s observed counts tend to show a higher unspliced contribution in some cells than others. In either case, a joint pattern is more informative than either count alone: a high \(U_{ig}\) could simply accompany high overall expression, while a high unspliced fraction describes a different balance.

Reading Counts Across Cells

A cell-by-gene count pair should be interpreted in context. The same unspliced count can mean something different when the spliced count and total count differ. Comparing \(U_{ig}\) alone can mistake high overall abundance for a shift toward intronic evidence. Comparing fractions without counts has the opposite danger: a fraction based on a handful of observations can look extreme by chance or because of measurement noise.

The gene-level summary also requires a choice about what is being averaged. Pooling counts across cells weights cells with more observations more heavily. Taking an unweighted mean of cell-specific fractions gives each eligible cell equal weight. These summaries answer different descriptive questions and should not be silently substituted for one another.

Assumptions, Diagnostics, and Alternatives

Interpreting these measurements requires several practical conditions. The gene annotation must identify relevant exons and introns, alignments must be sufficiently accurate, and the quantification procedure must classify ambiguous evidence consistently. For comparisons across cells, differences in sequencing depth, assay chemistry, RNA quality, and processing can affect the counts. A biological interpretation of the relative balance also assumes that technical differences in detecting intronic and exonic evidence are not driving the pattern.

Useful checks include examining the number of cells with nonzero counts, the distribution of \(U_{ig}+S_{ig}\), and whether a striking fraction is based on a very small denominator. Check how the pipeline handles reads that overlap feature boundaries and whether intronic evidence is captured by the assay at all. If intronic counts are nearly absent across the data, that may reflect the protocol or filtering rather than a universal absence of recent transcription. If apparent intronic signal is widespread in unexpected genes or cells, annotation, alignment, and contamination issues deserve attention.

Several alternatives are useful for different questions. An exon-only gene count can summarize expression evidence without separating the two categories, but it loses the distinction between intronic and spliced evidence. A combined count can provide a broader measure of gene-associated signal, but it likewise discards the balance. A pooled gene-level ratio is compact, while cell-specific ratios preserve cell-to-cell variation. None is universally best: choose the summary that matches the question and state how it was calculated.

Spliced and unspliced counts alone do not determine a transcription rate, a processing rate, or the direction of change over time. A high unspliced fraction may be consistent with recent transcription, but it can also reflect slower processing, differential degradation, or technical bias. Inferring a direction or a kinetic quantity requires additional assumptions and analysis; the count pair by itself is a measurement, not a complete dynamical explanation.

Common Mistake

A common mistake is to reason, “This cell has more unspliced counts, so the gene is being transcribed faster in this cell.” That conclusion does not follow from the count alone. A cell with more total RNA or greater sequencing depth can have more unspliced reads without a higher unspliced fraction; differences in processing, degradation, or capture can also change the count. Compare the unspliced and spliced evidence together, inspect the total and the denominator, and describe the result as a pattern in the observed counts unless further analysis supports a stronger interpretation.