Ambient RNA and Doublet Detection in Single-Cell RNA Sequencing

  • Ambient RNA contamination and doublet formation are two important technical challenges in single-cell RNA sequencing (scRNA-seq). Both can distort gene-expression profiles, affect cell-type identification and introduce errors into downstream computational analyses. Ambient RNA consists of extracellular RNA released from damaged or lysed cells that becomes captured during single-cell library preparation. Doublets occur when two or more cells are captured and assigned the same cell barcode, producing a mixed gene-expression profile. Identifying and addressing these problems is essential for obtaining reliable single-cell transcriptomic data.
  • Ambient RNA contamination occurs when RNA molecules released from cells enter the surrounding solution during tissue processing, cell isolation or library preparation. Cells can be damaged by mechanical disruption, enzymatic digestion, temperature changes, prolonged handling or unsuitable storage conditions. As damaged cells release their RNA, the resulting extracellular RNA becomes part of the suspension containing intact cells. During droplet-based sequencing, this RNA may be captured by barcode-bearing beads even when the corresponding droplet contains a genuine cell. Consequently, the expression profile assigned to that cell may contain transcripts that did not originate from it.
  • Ambient RNA contamination can affect several aspects of single-cell analysis. For example, highly abundant transcripts released from one cell population may appear in other cell types, creating false-positive gene-expression signals. This can make a gene appear to be expressed in a cell population where it is not normally active. It may also complicate cell-type annotation, differential expression analysis, gene regulatory network construction and the identification of rare cell populations. In some cases, ambient RNA can make unrelated cell populations appear more similar than they really are.
  • The composition of ambient RNA generally reflects the cells that have been damaged or lysed during sample preparation. Abundant cell populations may contribute disproportionately to the background RNA pool. For example, haemoglobin transcripts released from damaged red blood cells can contaminate other cellular profiles, while highly expressed tissue-specific transcripts may appear in unrelated cell populations. In pancreatic tissue, transcripts from abundant secretory cell populations may contribute to background contamination. The specific genes involved depend on tissue composition, cell fragility and experimental conditions.
  • Ambient RNA contamination can be investigated by examining gene-expression patterns across cell-associated barcodes and empty droplets. Empty droplets do not contain intact cells but may contain extracellular RNA captured during library preparation. Their expression profiles can provide information about the background RNA pool. Researchers can compare the expression of highly abundant marker genes across different cell populations and determine whether unexpected expression patterns may be explained by contamination. However, the presence of a transcript in multiple cell types does not automatically indicate contamination because some genes are genuinely expressed across different populations.
  • Several computational tools have been developed to estimate and correct ambient RNA contamination. EmptyDrops, available through the DropletUtils package in Bioconductor, identifies cell-containing droplets by comparing their RNA profiles with the ambient background. It is primarily a cell-calling method rather than a general ambient RNA removal tool. SoupX estimates the composition of the ambient RNA pool and adjusts cell-associated expression counts to reduce contamination. DecontX, available through the celda package, uses a statistical mixture model to estimate contamination in individual cell profiles. CellBender provides probabilistic methods for removing background RNA and identifying cell-associated signals from droplet-based sequencing data. These tools use different assumptions and approaches, so their outputs should be evaluated carefully.
  • Ambient RNA contamination can also be reduced through careful experimental procedures. Researchers should use fresh and appropriately preserved tissue, minimize unnecessary handling, optimize mechanical and enzymatic dissociation, maintain suitable temperatures and remove dead cells when appropriate. Filtration and centrifugation can help eliminate cellular debris, while cell viability assessment can identify samples with substantial cell damage. However, aggressive tissue processing may itself damage fragile cell populations or alter gene expression. Experimental optimization is therefore important for balancing cell recovery and sample quality.
  • Doublets are another major source of technical error in single-cell RNA sequencing. A doublet forms when two cells are captured together and receive the same cell barcode. In droplet-based platforms, this may occur when more than one cell enters a droplet containing a barcode-bearing bead. Because the resulting RNA molecules share the same barcode, sequencing analysis may interpret them as originating from one cell. A doublet can contain transcripts from two different cell types, producing a mixed expression profile that does not accurately represent either individual cell.
  • Doublets are generally classified into two broad categories. Homotypic doublets contain two cells of the same or a very similar cell type, whereas heterotypic doublets contain cells from different cell types. Heterotypic doublets are often easier to identify because they may express marker genes associated with distinct populations. For example, a barcode containing both T-cell markers and epithelial-cell markers may represent a heterotypic doublet. Homotypic doublets are more difficult to detect because their expression profiles may resemble those of larger or more transcriptionally active single cells.
  • Doublets can produce misleading biological conclusions. They may form artificial clusters, create false transitional cell populations, distort cell-type proportions and introduce incorrect marker genes. They can also affect differential expression analysis and gene regulatory network inference. In developmental studies, a doublet containing cells from two different developmental stages may be mistakenly interpreted as an intermediate cell state. In tumour studies, doublets involving malignant and immune cells may produce misleading expression profiles that complicate the interpretation of tumour microenvironments.
  • The probability of doublet formation is influenced by cell concentration, loading concentration, the number of cells processed and the characteristics of the sequencing platform. Higher cell loading generally increases the risk of multiple cells entering the same droplet. The expected doublet rate is platform- and loading-dependent and should be estimated using experimental guidance and appropriate controls. Doublet detection should therefore consider both computational evidence and the expected capture rate rather than relying only on gene or UMI counts.
  • Computational doublet detection methods generally use expression profiles to identify cells that resemble combinations of two or more individual cells. Scrublet generates artificial doublets by combining observed cell profiles and compares these simulated profiles with the real dataset. DoubletFinder uses artificial doublets and neighbourhood-based analysis to identify likely doublet-containing profiles. scDblFinder uses a computational classification approach based on simulated doublets and expression characteristics. SOLO uses a deep-learning framework to distinguish singlets from doublets. Other approaches, including DoubletDecon, use different strategies to identify mixed expression profiles.
  • A common strategy is to generate artificial doublets by combining the expression counts of two randomly selected cells. The resulting simulated profiles are compared with observed cells to identify those that resemble doublets. This approach is useful because genuine doublets are not usually known in advance. However, computational detection depends on the quality and diversity of the reference cell population, and simulated doublets may not perfectly reproduce all biological and technical characteristics of real doublets.
  • Doublet detection can also use biological information. Researchers can examine whether a cell expresses mutually incompatible marker genes, whether it contains unusually high gene or UMI counts, and whether its expression profile is inconsistent with known cell identities. For example, simultaneous expression of distinct lineage-specific markers may suggest a doublet, although genuine transitional states, cell-cell interactions and biological co-expression must also be considered. High gene counts alone are insufficient evidence because some genuine cells naturally contain more RNA than others.
  • Ambient RNA correction and doublet detection address different problems and should not be treated as interchangeable procedures. Ambient RNA correction attempts to reduce the influence of extracellular transcripts on cell-associated expression profiles. Doublet detection identifies barcodes that may represent more than one captured cell. A dataset may contain both problems simultaneously, and correcting one does not automatically resolve the other. A suitable analysis workflow should therefore assess cell quality, identify genuine cell-containing barcodes, estimate ambient contamination, detect potential doublets and evaluate the corrected expression matrix before downstream analysis.
  • A typical computational workflow begins with sequencing quality assessment using FastQC and MultiQC, followed by barcode processing, read alignment and gene quantification using tools such as Cell Ranger, STARsolo or alevin-fry. Cell calling can then identify barcodes likely to represent genuine cells. Ambient RNA contamination may be evaluated or corrected using SoupX, DecontX or CellBender, while doublets can be assessed using Scrublet, DoubletFinder, scDblFinder or SOLO. The resulting data can be analysed using Seurat, Scanpy or Bioconductor packages for normalization, dimensionality reduction, clustering, cell-type annotation and differential expression.
  • Quality assessment should be performed before and after correction to determine whether the methods have improved the data without removing genuine biological signals. Researchers can compare gene-expression distributions, marker expression, cell-type assignments, clustering patterns and the number of retained cells. It is also important to document the software versions, parameter settings, filtering thresholds and number of cells removed or corrected. Overcorrection can suppress genuine low-level expression, while overly aggressive doublet removal can eliminate legitimate cells with unusual transcriptional profiles.
  • In single-nucleus RNA sequencing (snRNA-seq), ambient RNA and doublet detection remain important, although the technical characteristics differ from whole-cell sequencing. Nuclei isolation can release RNA from damaged cells and contribute to background contamination. Nuclear datasets may also contain intronic reads and have different expression distributions from whole-cell datasets. Doublets can occur when multiple nuclei are captured under one barcode. Therefore, QC procedures and computational thresholds should be adapted to nuclear data rather than directly copied from conventional scRNA-seq workflows.
  • In animal genomics and breeding research, ambient RNA and doublet detection are particularly important when analysing complex tissues such as muscle, mammary gland, liver, ovary, testis and immune organs. Contamination from abundant cell populations may produce misleading expression of genes associated with growth, metabolism, immunity or reproduction. Doublets may create artificial cell populations and distort the identification of cell types involved in economically important traits. Reliable correction and detection support more accurate investigations of cellular mechanisms associated with muscle development, milk production, fertility, feed efficiency and disease resistance. Integration with genomic variation, quantitative trait loci and phenotypic information can further connect cell-specific gene expression with complex animal traits.
  • Ambient RNA contamination and doublet formation are unavoidable considerations in many single-cell sequencing experiments, but their effects can be assessed and reduced through appropriate experimental design, quality control and computational analysis. Ambient RNA introduces extracellular transcripts into cell-associated profiles, whereas doublets combine the expression signals of multiple cells under one barcode. Understanding the distinction between these problems and applying suitable tools helps improve cell-type identification, gene-expression quantification and the biological interpretation of single-cell and single-nucleus transcriptomic datasets.
Author:

Leave a Reply

Your email address will not be published. Required fields are marked *