- Quality control (QC) and cell filtering are essential steps in single-cell RNA sequencing (scRNA-seq) that ensure reliable gene-expression measurements and accurate identification of cellular populations. Single-cell experiments generate large datasets containing thousands or millions of cell-associated barcodes, but not every barcode represents a healthy, intact cell. Some may originate from empty droplets, damaged cells, dying cells, cell fragments, ambient RNA or multiple cells captured together. Technical problems during tissue preparation, cell isolation, RNA capture, reverse transcription, amplification and sequencing can also affect data quality. Quality control identifies these problems, while cell filtering removes or flags low-quality observations before downstream analysis.
- The primary objective of quality control is to distinguish biologically meaningful gene-expression profiles from technical noise and unreliable measurements. A typical single-cell RNA-seq dataset contains a cell-by-gene expression matrix in which rows represent individual cells, columns represent genes and entries represent detected RNA molecules or sequencing counts. Before normalization, clustering and differential expression analysis, researchers examine several quality metrics to assess the reliability of each cell. The most commonly evaluated metrics include the number of detected genes, total RNA or UMI counts, mitochondrial gene expression, ribosomal gene expression, the proportion of reads assigned to genes, and the presence of potential doublets or contaminated profiles.
- The number of detected genes per cell is an important indicator of transcriptome complexity. A cell with very few detected genes may contain insufficient RNA because it was damaged, poorly captured or inadequately sequenced. However, the expected number of detected genes varies considerably among cell types. Large or transcriptionally active cells may express more genes than small or relatively inactive cells. Therefore, unusually low gene counts can indicate poor-quality cells, while unusually high gene counts may suggest doublets or multiplets. These observations should be interpreted in the context of the tissue, organism, sequencing platform and experimental design.
- The total number of unique molecular identifiers (UMIs) per cell is another important quality metric in UMI-based protocols. UMIs help estimate the number of original RNA molecules captured from each cell. A low UMI count may indicate low RNA content, inefficient capture, cell damage or insufficient sequencing depth. An unusually high UMI count may indicate a doublet, a large or highly active cell, or an exceptionally high level of transcript capture. UMI counts are therefore useful for identifying unusual observations, but they should not be used as a universal measure of cell quality.
- Mitochondrial gene expression is widely used to identify potentially damaged or stressed cells. Mitochondrial genes encode components of the mitochondrial respiratory machinery, and their RNA contribution can become relatively high when cytoplasmic RNA is lost from damaged cells. The percentage of mitochondrial counts is commonly calculated as the proportion of total counts assigned to mitochondrial genes. Cells with unusually high mitochondrial RNA percentages may be compromised and may have unreliable gene-expression profiles. However, some healthy cell types naturally show relatively high mitochondrial activity, so mitochondrial thresholds must be selected according to the biological system and tissue being studied.
- The percentage of mitochondrial RNA can be calculated as:
- Percentage of mitochondrial counts = (Mitochondrial RNA counts / Total RNA counts) × 100
- Here, mitochondrial RNA counts represent the total counts assigned to mitochondrial genes, while total RNA counts represent all counted genes in the cell. The resulting percentage indicates the relative contribution of mitochondrial transcripts to the cell’s expression profile.
- Ribosomal gene expression can also provide useful information about cellular quality and transcriptional activity. Ribosomal RNA and ribosomal protein genes are involved in protein synthesis, and their expression may vary across cell types and physiological conditions. Unusually high ribosomal gene proportions can sometimes indicate changes in cellular state or technical effects. However, high ribosomal expression is not necessarily evidence of poor quality. It should be interpreted alongside other metrics rather than used as an independent reason for removing cells.
- For single-nucleus RNA sequencing (snRNA-seq), quality control requires some additional considerations because the method measures RNA obtained from isolated nuclei rather than whole cells. Nuclear datasets often contain a higher proportion of intronic reads because they capture unspliced and partially processed pre-mRNAs. They also commonly show lower mitochondrial RNA proportions than whole-cell datasets. Consequently, quality-control thresholds developed for conventional scRNA-seq should not automatically be applied to snRNA-seq. Researchers should evaluate nuclear RNA content, detected genes, UMI counts, intronic read fractions, ambient RNA contamination and evidence of damaged or aggregated nuclei.
- A further challenge is the presence of empty droplets and background RNA. During droplet-based sequencing, some droplets contain no intact cell but may still capture RNA released from damaged or lysed cells. These droplets can produce barcode-associated sequencing reads even though they do not represent genuine cellular transcriptomes. Ambient RNA can also enter droplets containing real cells, creating artificial expression signals. This contamination may cause genes from abundant cell types to appear in other cell populations. Computational methods such as EmptyDrops, SoupX and DecontX help identify likely empty droplets or estimate and correct ambient RNA contamination. SoupX and DecontX are designed to address contamination in cell-associated expression profiles, while EmptyDrops helps distinguish genuine cell-containing droplets from background.
- Doublets and multiplets occur when two or more cells are captured under the same barcode. Their expression profiles may contain genes associated with different cell types, potentially producing misleading clusters or artificial cell populations. Doublets can arise naturally during cell capture and are particularly relevant in high-throughput droplet-based experiments. They may be identified through unusually high gene or UMI counts, mixed expression signatures, computational modelling and comparison with expected cell-capture rates. Common doublet-detection tools include Scrublet, DoubletFinder, scDblFinder and SOLO. These tools use different computational strategies, so their results should be interpreted as evidence of potential doublets rather than unquestionable classifications.
- Cell filtering involves removing or flagging barcodes that fail predefined quality criteria. Common filtering approaches include excluding cells with very low gene counts, unusually low UMI counts, excessive mitochondrial RNA proportions, abnormal ribosomal proportions or evidence of doublet formation. Researchers may also remove cells with high levels of ambient RNA contamination or poor sequencing quality. However, filtering should be conservative and biologically informed. Excessively strict thresholds can remove genuine cell populations, especially rare cells, small cells, fragile cell types or cells with naturally low RNA content. Conversely, permissive filtering may retain damaged cells and technical artefacts that interfere with downstream analysis.
- A typical quality-control workflow begins with the examination of raw sequencing reads and sequencing-library quality. FastQC provides information about base quality, sequence composition, adapter contamination and other sequencing characteristics, while MultiQC combines quality reports from multiple samples into a single summary. After barcode processing, alignment and gene quantification, researchers use cell-level quality metrics to evaluate the resulting expression matrix. Common tools for downstream QC include Seurat, Scanpy, scater and scuttle. Seurat and Scanpy provide functions for calculating quality metrics, visualizing distributions and filtering cells. Bioconductor packages such as scater and scuttle provide additional tools for quality assessment and data handling.
- Visualization is an important part of quality control because it helps researchers understand the distribution of quality metrics and identify unusual observations. Violin plots, histograms, scatter plots and box plots can reveal differences in detected genes, UMI counts and mitochondrial percentages among cells or samples. For example, plotting the number of detected genes against total UMI counts may help identify cells with unusually high or low transcript complexity. Comparing mitochondrial percentages across samples can reveal differences in tissue quality or sample preparation. These visualizations also help researchers select appropriate filtering thresholds rather than relying on arbitrary fixed values.
- Quality control should also consider differences between biological samples and experimental batches. Variations in tissue collection, storage time, dissociation procedures, cell viability, library preparation and sequencing depth can produce technical differences among samples. Batch effects may influence gene-expression patterns and create artificial clusters that do not reflect biological variation. Researchers should therefore examine QC metrics separately for each sample and experimental batch before combining datasets. Batch correction methods such as Harmony, Seurat integration, fastMNN and scVI may be useful later in the analysis, but they should not replace proper initial quality control.
- After filtering, the retained cells are generally used for normalization, identification of highly variable genes, dimensionality reduction, clustering and cell-type annotation. The quality of these downstream analyses depends strongly on the reliability of the initial expression matrix. Poor-quality cells can introduce noise, distort gene-expression distributions, affect principal component analysis and create misleading clusters. Incorrect removal of valid cells can be equally problematic because it may reduce biological diversity and eliminate rare populations of interest. Therefore, quality control is not simply a technical cleaning step; it is an important part of experimental interpretation.
- In animal genomics and breeding research, quality control and cell filtering are particularly important when analysing tissues associated with economically important traits. Studies of muscle, mammary gland, ovary, testis, liver, adipose tissue and immune organs may involve cell types with very different sizes, RNA contents and metabolic activities. For example, mitochondrial gene expression may be naturally high in metabolically active cells, while some specialised cell types may contain relatively little RNA. Applying identical filtering thresholds across all tissues could therefore remove biologically important populations. Careful QC supports the reliable identification of cell populations associated with muscle development, milk production, fertility, feed efficiency, immune responses and disease resistance.
- Quality control and cell filtering are fundamental to producing reliable single-cell transcriptomic data. By evaluating gene counts, UMI counts, mitochondrial expression, sequencing quality, ambient RNA contamination and doublet formation, researchers can identify technical artefacts and retain high-quality cellular profiles. Appropriate filtering improves the reliability of downstream computational analyses, but thresholds must be adapted to the tissue, organism, sequencing method and research objectives. A combination of statistical assessment, visualization, biological knowledge and independent validation provides the strongest foundation for interpreting single-cell and single-nucleus RNA sequencing data.