Single-Cell RNA Sequencing Quality Control: A Practical Guide to Filtering and Metrics

Single-cell RNA sequencing (scRNA-seq) generates transcriptome-wide expression profiles from individual cells, but the data arriving from the sequencer contain technical artifacts that must be identified and removed before biological interpretation. Quality control (QC) is the process of distinguishing high-quality cell barcodes from empty droplets, damaged cells, doublets, and ambient RNA contamination. This guide provides a practical framework for applying QC metrics, setting filtering thresholds, and troubleshooting common failures in scRNA-seq datasets.

The intended reader is a researcher, analyst, or student who has generated or downloaded scRNA-seq count matrices and needs to make defensible filtering decisions. The guidance assumes familiarity with basic R or Python programming and the structure of count matrices produced by platforms such as 10X Genomics Chromium or BD Rhapsody. The focus is on droplet-based whole-cell and single-nucleus data, with attention to the biological and technical factors that influence QC metric interpretation.

Why Quality Control Matters in Single-Cell RNA Sequencing

The fundamental premise of scRNA-seq is that each cell barcode corresponds to the transcriptome of one cell. In practice, this premise is violated by several classes of artifacts. Empty droplets capture cell-free RNA from the input solution, producing barcodes with low RNA content that can be mistaken for real cells with low transcriptional activity. Damaged or dying cells release their cytoplasmic RNA while retaining mitochondrial transcripts, creating barcodes with high mitochondrial read fractions. Doublets occur when two or more cells share a single barcode, producing expression profiles that are artificial mixtures of distinct cell types. Ambient RNA, often called the soup, consists of cell-free transcripts that contaminate every droplet and distort expression measurements.

The consequences of inadequate QC are substantial. Low-quality cells add noise that obscures rare cell populations and inflates apparent heterogeneity. Doublets can create false cell types that appear as intermediate states between genuine populations. Ambient RNA contamination shifts expression profiles toward highly expressed genes from abundant cell types, confounding differential expression analysis. A study of single-cell perturbation datasets found that uniform quality control pipelines were necessary to enable comparison and integration across experiments, highlighting the importance of consistent filtering practices [6].

The scale of the problem is documented in the literature. Droplet-based scRNA-seq assays capture cell-free RNA from the input solution, and this background contamination is ubiquitous with experiment-specific variations in composition and magnitude [7]. The isolation of cells or nuclei for sequencing releases contaminating RNA through cell damage and transcript leakage, making the identification of high-quality barcodes a critical analytical step [5]. These observations establish that QC is not an optional refinement but a required step in any scRNA-seq analysis.

Core Quality Control Metrics and Their Biological Basis

The standard QC metrics used across scRNA-seq pipelines are the number of unique molecular identifiers (UMIs), the number of detected genes, and the percentage of reads mapping to mitochondrial genes. Each metric reflects a different aspect of cell and library quality.

UMI Count and Gene Count

The UMI count estimates the total number of transcripts captured per cell. Low UMI counts indicate either empty droplets, cells with low transcriptional activity, or cells that were lost during library preparation. The gene count measures the number of distinct genes detected per cell. Cells with very low gene counts may be broken or dying, while cells with unusually high gene counts may be doublets.

The relationship between UMI count and gene count is informative. A cell with high UMI count but low gene count suggests amplification of a small number of transcripts, which can occur with degraded RNA. A cell with both high UMI and high gene counts may be a doublet, particularly if the expression profile combines markers from distinct cell types.

Mitochondrial Read Percentage

The percentage of reads mapping to mitochondrial genes is a widely used proxy for cell viability. Intact cells have a characteristic mitochondrial fraction that varies by tissue and cell type. Cells with elevated mitochondrial percentages are typically damaged, with cytoplasmic RNA lost through a compromised membrane while mitochondrial transcripts remain trapped. The valiDrops method uses data-adaptive thresholding on community-standard quality metrics and can predict and flag dead cells with high accuracy, demonstrating that mitochondrial content is a reliable indicator of cell state [5].

However, the mitochondrial percentage must be interpreted with biological context. Some cell types naturally have high mitochondrial content due to their metabolic demands. A study of skeletal muscle myofibers found that quality control metrics across different flow-gating strategies identified only a subset of cells as optimal, with a median read count of 239,252 and an average of 12,098 transcripts per cell [11]. Muscle cells are metabolically active and may have higher baseline mitochondrial fractions than other cell types. Similarly, a case report of monocytes in hepatitis B-related acute-on-chronic liver failure identified an inflammatory monocyte subpopulation with upregulated mitochondrial genes, indicating that mitochondrial gene expression can carry biological signal instead of only technical noise [14].

Intronic Fraction and Nuclear Markers

For single-nucleus RNA sequencing (snRNA-seq), the intronic fraction is an important QC metric. Nuclei contain unspliced precursor mRNA, so a high proportion of intronic reads confirms that the barcode corresponds to a nucleus instead of a whole cell or ambient RNA. The scQCenrich framework integrates intronic fraction and MALAT1 enrichment with canonical metrics to improve QC decisions for whole-cell data [13]. MALAT1 is a long noncoding RNA that is highly expressed in the nucleus, so its enrichment can help distinguish nuclear from cytoplasmic contamination.

Doublet Rate

Doublets are a predictable consequence of droplet-based encapsulation. The doublet rate increases with the number of cells loaded onto the microfluidic device. Computational doublet detection methods simulate doublets from the observed data and train classifiers to distinguish them from true single cells. The Solo method uses a semi-supervised deep learning approach that embeds cells with a variational autoencoder and appends a supervised classifier to identify doublets with greater accuracy than existing methods [20]. The DoubletCollection package integrates the installation and execution of eight doublet detection methods, providing a unified interface for benchmarking their performance [23].

A notable finding from plant scRNA-seq research is that computational doublet detection algorithms can misclassify a substantial fraction of true single cells. In a study comparing protoplast enrichment strategies, single-nucleotide polymorphism analysis of a mixed ecotype sample revealed that among cells identified as doublets by computational algorithms, two-thirds were likely to have been misclassified [19]. This observation underscores the importance of validating doublet calls with independent evidence, such as genotype information or known marker gene co-expression.

At a Glance: QC Metrics and Filtering Decisions

The following table summarizes the primary QC metrics, their biological interpretation, common filtering approaches, and caveats for interpretation.

The thresholds in this table are starting points, not universal rules. The pipeComp framework for pipeline comparison demonstrated that the optimal filtering strategy depends on the dataset and the downstream analysis goals [21]. A filtering threshold that works well for a homogeneous cell line may remove too many cells from a heterogeneous tissue sample, while a permissive threshold may retain excessive noise in a high-quality dataset.

Practical Workflow for Quality Control

The QC workflow proceeds through several stages, from raw data import to filtered count matrices ready for normalization and downstream analysis. The following sequence represents a standard approach that can be adapted to specific datasets and platforms.

Step 1: Import Data and Generate QC Metrics

The first step is to import the count matrix and compute QC metrics for each cell barcode. The singleCellTK package provides the SCTK-QC pipeline, which imports data from several single-cell platforms and preprocessing tools and includes steps for empty droplet detection, generation of standard QC metrics, prediction of doublets, and estimation of ambient RNA [10]. The pipeline can run on the command line, within the R console, on a cloud platform, or with an interactive graphical user interface.

The popsicleR package offers an interactive approach to pre-processing and QC analysis, starting from either the output files of the Cell Ranger pipeline from 10X Genomics or from a feature-barcode matrix of raw counts generated from any scRNA-seq technology [9]. The package integrates methods for the estimation of quality-control metrics, filtering of low-quality cells, data normalization, removal of technical and biological biases, and cell clustering and annotation.

Step 2: Identify and Remove Empty Droplets

Empty droplets contain ambient RNA but no cell. The barcode rank plot, which shows the UMI count for each barcode sorted in descending order, typically displays a sharp drop at the transition between real cells and empty droplets. The knee of this curve provides an initial threshold for cell calling.

Automated methods can improve on manual threshold selection. The valiDrops method initially filters barcodes using data-adaptive thresholding on community-standard quality metrics, then uses a clustering-based approach to identify barcodes with distinct biological signals [5]. This two-stage approach reduces the risk of removing real cells with low transcriptional activity while retaining the ability to flag dead cells.

Step 3: Filter Low-Quality Cells

After removing empty droplets, apply thresholds for UMI count, gene count, and mitochondrial percentage. The specific thresholds should be informed by the distribution of these metrics in the dataset. A common approach is to examine the distributions and set thresholds at visible inflection points, then verify that the retained cells form coherent clusters in downstream analysis.

The scQCenrich framework provides an interpretable multi-metric QC approach that integrates canonical metrics with intronic fraction, MALAT1 enrichment, and dissociation-stress features [13]. Across mouse brain, mouse heart, and lung cancer datasets, scQCenrich reduced over-filtering relative to conventional and model-based comparators while preserving coherent neuronal, erythroid, cardiomyocyte, and malignant-cell populations. In high-quality peripheral blood mononuclear cell data, the method remained conservative.

Step 4: Detect and Remove Doublets

Doublet detection should be performed after initial cell filtering, because the presence of low-quality cells can interfere with doublet classification. Run one or more computational doublet detection methods and compare their predictions. The DoubletCollection package provides a unified interface for executing and benchmarking eight doublet detection methods [23].

The expected doublet rate can be estimated from the number of cells loaded onto the microfluidic device. If the observed doublet rate substantially exceeds the expected rate, this may indicate problems with cell concentration or loading. Conversely, if the doublet detection method flags an implausibly high fraction of cells, the thresholds may be too aggressive.

Step 5: Estimate and Correct Ambient RNA

Ambient RNA contamination affects all droplet-based scRNA-seq data. The SoupX method quantifies the extent of contamination and estimates background-corrected cell expression profiles that integrate with existing downstream analysis tools [7]. Applying SoupX to several datasets using multiple droplet sequencing technologies improved biological interpretation of otherwise misleading data and improved quality control metrics.

Ambient RNA correction is particularly important for datasets with strong expression of a small number of highly abundant genes, such as hemoglobin in blood samples or ribosomal genes in many tissues. The contamination fraction should be estimated and reported, and the decision to apply correction should be documented.

Step 6: Assess Data Quality After Filtering

After filtering, examine the distribution of QC metrics in the retained cells. The filtered dataset should show a relatively uniform distribution of UMI counts and gene counts, with mitochondrial percentages within the expected range for the tissue. Visualize the data with dimensionality reduction and clustering to confirm that the retained cells form biologically coherent populations.

The IBRAP pipeline provides integrated benchmarking of scRNA-seq analytical pipelines, allowing users to compare results across different combinations of preprocessing, quality control, normalization, dimensionality reduction, integration, and clustering steps [12]. The optimal pipeline combination depends on the individual samples and studies, so benchmarking is a necessary step for rigorous analysis.

Setting Filtering Thresholds: Decision Criteria

The choice of filtering thresholds is the most consequential decision in scRNA-seq QC. Thresholds that are too aggressive remove real biological variation, while thresholds that are too permissive retain technical noise. The following decision criteria provide a structured approach.

Data-Driven Threshold Selection

Examine the distribution of each QC metric and identify natural breakpoints. The barcode rank plot provides a clear visual signal for the cell versus empty droplet boundary. For UMI count and gene count, look for a shoulder or inflection point in the distribution. For mitochondrial percentage, look for a distinct population of cells with elevated values that separates from the main distribution.

The valiDrops method uses data-adaptive thresholding on community-standard quality metrics, which adjusts thresholds based on the characteristics of each dataset instead of applying fixed values [5]. This approach is preferable to fixed thresholds because the optimal values vary by tissue, protocol, and sequencing depth.

Biological Context Considerations

The tissue type and cell composition of the sample should inform threshold selection. Tissues with metabolically active cells, such as muscle or heart, may have higher baseline mitochondrial percentages. Tissues with large cells, such as hepatocytes or neurons, may have higher UMI counts per cell. The expected cell type composition should be considered when evaluating whether the filtered dataset retains the expected populations.

A study of pancreatic neuroendocrine tumors used publicly available scRNA-seq data and resolved the cell population into ten transcriptionally distinct clusters, including tumor cells, endothelial cells, fibroblasts, immune cells, and podocyte-like cells [18]. The dominant tumor cell population constituted 67 percent of the cells, and metabolic state annotation identified three states distributed differentially across tumor sub-clusters. This type of expected composition provides a reference for evaluating whether filtering has removed an entire cell population.

Sensitivity Analysis

Test the impact of different thresholds on downstream results. Run the analysis with permissive, moderate, and stringent thresholds and compare the clustering results, marker gene detection, and differential expression outcomes. If the biological conclusions are stable across threshold choices, the analysis is robust to filtering decisions. If the conclusions change dramatically, the thresholds require closer scrutiny.

The pipeComp framework was designed to handle interactions between analysis steps and relies on multi-level evaluation metrics [21]. The framework covers common methods of filtering, doublet detection, normalization, feature selection, denoising, dimensionality reduction, and clustering, allowing systematic evaluation of how each step affects the final results.

Records and Measurements for Quality Control

Documentation of QC decisions is essential for reproducibility and for comparison across datasets. The following records should be maintained for each scRNA-seq dataset.

Pre-Filtering Metrics

Record the total number of barcodes, the number of barcodes identified as cells, and the number of barcodes removed at each filtering step. Record the distribution of UMI counts, gene counts, and mitochondrial percentages before and after filtering. These numbers provide the basis for evaluating whether the filtering was appropriate.

Filtering Parameters

Document the exact thresholds applied for each QC metric and the rationale for each threshold. If data-adaptive methods were used, record the parameters and the version of the software. If thresholds were adjusted based on biological considerations, document the reasoning.

Post-Filtering Quality Assessment

Record the number of cells retained, the median UMI count, the median gene count, and the median mitochondrial percentage after filtering. Record the predicted doublet rate and the estimated ambient RNA contamination fraction. These metrics provide a summary of the final data quality.

Software and Version Information

Record the versions of all software packages used for QC, including the alignment tool, the cell calling method, the doublet detection method, and the ambient RNA correction method. Software versions affect results, and reproducibility requires precise documentation.

The FAIR Guiding Principles provide a framework for data management that applies to scRNA-seq QC records [4]. Making QC records findable, accessible, interoperable, and reusable supports the broader goal of reproducible research. The NIH Genomic Data Sharing Policy provides additional requirements for data sharing that may apply to federally funded research [3].

Common Failure Patterns in scRNA-seq QC

Several failure patterns recur across scRNA-seq datasets. Recognizing these patterns allows analysts to diagnose problems and adjust their approach.

Over-Filtering

Over-filtering occurs when thresholds remove too many cells, including real biological populations. This is most common when fixed thresholds are applied without examining the data distribution. The scQCenrich framework was developed specifically to address over-filtering, demonstrating that conventional and model-based comparators removed coherent cell populations that should have been retained [13].

Signs of over-filtering include the loss of expected cell types, a sudden drop in cell number at a threshold boundary, and the absence of low-expression cell populations that are known to exist in the tissue. If over-filtering is suspected, relax the thresholds and examine whether the additional cells form coherent clusters with distinct marker expression.

Under-Filtering

Under-filtering occurs when thresholds are too permissive, retaining empty droplets, damaged cells, and doublets. Signs of under-filtering include a large number of cells with very low UMI counts, a population of cells with high mitochondrial percentages, and clusters that express markers from multiple cell types.

Under-filtering is often detected during downstream analysis when clusters cannot be annotated or when differential expression results are dominated by technical artifacts. The solution is to revisit the QC thresholds and apply more stringent filtering.

Batch Effects Confounded with Quality

When multiple samples are processed in different batches, quality differences between batches can be confounded with biological differences. A sample with lower sequencing depth will have lower UMI counts and gene counts, which can cause its cells to cluster separately from other samples even if the biological composition is identical.

The scPerturb resource applied uniform quality control pipelines across 44 publicly available single-cell perturbation datasets and harmonized feature annotations to facilitate comparison and integration [6]. This approach demonstrates the value of consistent QC across batches. When batch effects are confounded with quality, consider applying sample-specific thresholds or using integration methods that account for quality differences.

Doublet Misclassification

Computational doublet detection methods can misclassify true single cells as doublets, particularly for cells with extreme transcriptional activity or unusual gene expression profiles. The plant scRNA-seq study found that two-thirds of cells identified as doublets by computational algorithms were likely misclassified based on single-nucleotide polymorphism analysis [19].

To mitigate doublet misclassification, validate doublet calls with independent evidence. Genotype information, when available, provides the most reliable validation. In the absence of genotype data, examine whether the predicted doublets express mutually exclusive marker genes from distinct cell types.

Ambient RNA Dominance

In samples with high ambient RNA contamination, the contamination can dominate the expression profiles of low-RNA cells, making them appear similar to the dominant cell type in the sample. This is particularly problematic in blood samples where hemoglobin transcripts are abundant in the ambient RNA.

The SoupX method quantifies the extent of contamination and estimates background-corrected cell expression profiles [7]. Applying ambient RNA correction before downstream analysis can substantially improve the biological interpretation of otherwise misleading data.

Limitations of Quality Control Metrics

QC metrics are imperfect proxies for cell quality, and their interpretation requires awareness of their limitations.

Mitochondrial Percentage Is Not a Universal Viability Marker

The assumption that high mitochondrial percentage indicates cell damage does not hold for all cell types. Cells with high metabolic activity, such as cardiomyocytes and skeletal muscle fibers, may have naturally elevated mitochondrial content. A study of skeletal muscle myofibers found that only a subset of cells met optimal quality criteria, with a median read count of 239,252 and an average of 12,098 transcripts per cell [11]. Applying a stringent mitochondrial threshold to such cells would remove biologically relevant populations.

Mitochondrial gene expression can also carry biological signal. The case report of monocytes in HBV-ACLF identified an inflammatory monocyte subpopulation with upregulated mitochondrial genes and enrichment of secreted factors associated with platelet activation and systemic inflammation [14]. The proportion of this subpopulation decreased with treatment, indicating that mitochondrial gene expression reflected disease state instead of technical artifact.

UMI and Gene Counts Reflect Biology as Well as Quality

Cells in different states have different transcriptional activity. Quiescent cells, circulating immune cells, and cells in G0 phase have lower UMI counts than actively proliferating or metabolically active cells. Applying a uniform UMI threshold across a heterogeneous tissue can remove real cell populations.

The normalization of single-cell RNA-seq data is an important step that interacts with QC decisions [8]. Data-driven metrics can be used to rank normalization methods and select the best performers, but the choice of normalization can have a profound impact on the results. QC filtering and normalization should be considered together instead of as independent steps.

Doublet Detection Methods Have Variable Performance

The performance of doublet detection methods varies by dataset and by the proportion of doublets present. The pipeComp benchmark found that tools operate differentially according to the data types and complexity [21]. No single doublet detection method outperforms all others in all datasets.

The protocol for benchmarking eight doublet detection methods using the DoubletCollection package provides a systematic approach to evaluating doublet detection performance [23]. Running multiple methods and comparing their predictions can identify consensus doublet calls that are more reliable than any single method.

Ambient RNA Correction Has Limitations

Ambient RNA correction methods estimate the contamination fraction and adjust expression profiles accordingly, but the correction is imperfect. The estimation of the contamination fraction relies on genes that are not expected to be expressed in the cell population, and the choice of these genes affects the correction.

The SoupX method demonstrates that contamination from cell-free RNA is ubiquitous with experiment-specific variations in composition and magnitude [7]. The method improves quality control metrics and biological interpretation, but the corrected profiles are estimates instead of exact measurements.

Safety and Regulatory Context for Data Handling

scRNA-seq data often contain sensitive information about human subjects. The handling of these data is subject to institutional and governmental policies.

Genomic Data Sharing

The NIH Genomic Data Sharing Policy establishes expectations for the sharing of genomic data generated through NIH-funded research [3]. The policy covers data deposition, data access, and data security requirements. Researchers working with human scRNA-seq data should be familiar with the policy requirements that apply to their research.

Data Repositories

The NCBI provides data resources for the deposition and retrieval of genomic data, including single-cell sequencing data [2]. The EMBL-EBI Training program offers educational resources on data management and analysis [1]. These resources support the responsible handling and sharing of scRNA-seq data.

Reproducibility and Documentation

The FAIR Guiding Principles provide a framework for making data findable, accessible, interoperable, and reusable [4]. Applying these principles to scRNA-seq QC records supports reproducibility and enables other researchers to evaluate the quality of published datasets.

Professional Escalation Criteria

Some QC problems require consultation with a bioinformatics specialist or a statistician. The following situations warrant escalation.

Persistent Batch Effects

If batch effects cannot be resolved through QC filtering, normalization, or integration methods, consult a specialist. Persistent batch effects can indicate problems with sample processing, library preparation, or sequencing that require experimental changes instead of computational fixes.

Unexpected Cell Type Composition

If the filtered dataset lacks expected cell types or contains unexpected populations, consult a specialist before proceeding with downstream analysis. The unexpected composition may indicate a biological finding, a technical artifact, or a problem with cell annotation.

Conflicting Doublet Detection Results

If multiple doublet detection methods produce conflicting results, consult a specialist to determine the appropriate interpretation. Conflicting results may indicate that the doublet rate is unusually high or that the methods are not appropriate for the data type.

Inability to Reproduce Published Results

If QC metrics or filtering results cannot be reproduced from published data, consult a specialist. The inability to reproduce results may indicate differences in software versions, parameter settings, or data processing steps.

Frequently Asked Questions

What is the difference between a cell barcode and a cell?

A cell barcode is a unique nucleotide sequence that identifies the droplets or wells in a single-cell experiment. Each barcode is intended to correspond to one cell, but in practice, barcodes can contain zero cells (empty droplets), one cell, or multiple cells (doublets). Quality control is the process of determining which barcodes contain high-quality single cells.

Why is the mitochondrial percentage used as a quality control metric?

The mitochondrial percentage reflects the fraction of reads mapping to mitochondrial genes. Cells with damaged membranes lose cytoplasmic RNA while retaining mitochondrial transcripts, leading to elevated mitochondrial percentages. However, some cell types naturally have high mitochondrial content, so the threshold must be interpreted with biological context.

What is ambient RNA and how does it affect single-cell data?

Ambient RNA is cell-free RNA present in the input solution that is captured by droplets along with cellular RNA. This contamination is ubiquitous in droplet-based experiments and can distort expression profiles, particularly for cells with low RNA content. Methods such as SoupX estimate and correct for ambient RNA contamination [7].

How do I choose the right filtering thresholds for my dataset?

Examine the distribution of QC metrics in your dataset and identify natural breakpoints. Use data-adaptive methods when available, and validate threshold choices with sensitivity analysis. Consider the biological context of your tissue and the expected cell type composition.

What is a doublet and why is it a problem?

A doublet is a barcode that contains two or more cells. Doublets produce expression profiles that are artificial mixtures of distinct cell types, which can create false cell populations and confound differential expression analysis. Computational doublet detection methods simulate doublets and train classifiers to identify them [20].

Should I use fixed thresholds or data-adaptive thresholds?

Data-adaptive thresholds are generally preferable because the optimal values vary by tissue, protocol, and sequencing depth. The valiDrops method uses data-adaptive thresholding on community-standard quality metrics [5]. Fixed thresholds can be used as starting points but should be adjusted based on the data distribution.

How do I know if I have filtered too many cells?

If the filtered dataset lacks expected cell types or if the cell number drops sharply at a threshold boundary, over-filtering may have occurred. Relax the thresholds and examine whether the additional cells form coherent clusters with distinct marker expression.

What should I report about quality control in my methods section?

Report the number of cells before and after filtering, the thresholds applied for each QC metric, the software and versions used, the predicted doublet rate, and the estimated ambient RNA contamination fraction. This information allows other researchers to evaluate the quality of your data and reproduce your analysis.

Related Bioinformatics Guides

• Single-Cell RNA Sequencing: From Bulk to Resolution
• Master Guide: Single-Cell RNA Sequencing Bioinformatics Workflows
• Single-Cell RNA-Seq Analysis Pipelines for Veterinary Immunology
• Single-cell RNA-seq Trajectory Inference and Cell Lineage Tracing
• Single-Cell RNA-seq Clustering and Cell-Type Annotation Pipelines

References and Further Reading

• EMBL-EBI Training. European Bioinformatics Institute.
• NCBI Data Resources. National Center for Biotechnology Information.
• Genomic Data Sharing Policy. National Institutes of Health.
• The FAIR Guiding Principles. Scientific Data.
• Automatic quality control of single-cell and single-nucleus RNA-seq using valiDrops.. NAR genomics and bioinformatics, 2023.
• scPerturb: harmonized single-cell perturbation data.. Nature methods, 2024.
• SoupX removes ambient RNA contamination from droplet-based single-cell RNA sequencing data.. GigaScience, 2020.
• Normalization of Single-Cell RNA-Seq Data.. Methods in molecular biology (Clifton, N.J.), 2021.
• popsicleR: A R Package for Pre-processing and Quality Control Analysis of Single Cell RNA-seq Data.. Journal of molecular biology, 2022.
• Comprehensive generation, visualization, and reporting of quality control metrics for single-cell RNA sequencing data.. Nature communications, 2022.
• Single cell RNA-seq analysis of the flexor digitorum brevis mouse myofibers.. Skeletal muscle, 2021.
• IBRAP: integrated benchmarking single-cell RNA-sequencing analytical pipeline.. Briefings in bioinformatics, 2023.
• ScQCenrich enables multi-metric quality control for single-cell RNA sequencing.. 2026.
• Case Report: dynamic monocyte reprogramming during ALSS therapy in type B HBV-ACLF revealed by single-cell transcriptomics.. 2026.
• Integrative analysis of bulk and single-nucleus transcriptomes suggests proteostasis- and metabolism-related alterations in the right ventricular outflow tract of non-syndromic Tetralogy of Fallot.. 2026.
• Stepwise Protocol for Alternative Splicing Analysis in Single-Cell SMART-Seq2 RNA-Seq Data. 2026.
• MitoPerturb-Seq identifies gene-specific single-cell responses to mitochondrial DNA depletion and heteroplasmy.. 2026.
• Mapping metabolic reprogramming dynamics across pancreatic neuroendocrine tumor cell differentiation at single-cell transcriptomic resolution.. 2026.
• Benchmarking plant single cell RNA-sequencing sample processing strategies.. 2026.
• Solo: Doublet Identification in Single-Cell RNA-Seq via Semi-Supervised Deep Learning.. Cell Systems, 2020.
• pipeComp, a general framework for the evaluation of computational pipelines, reveals performant single cell RNA-seq preprocessing tools. Genome Biology, 2020.
• scReady - an automated and accessible pipeline for single-cell RNA-Seq preprocessing: Empowering novice bioinformaticians. Wellcome Open Research, 2026.
• Protocol for executing and benchmarking eight computational doublet-detection methods in single-cell RNA sequencing data analysis. STAR Protocols, 2021.
• dupRadar: A Bioconductor package for the assessment of PCR artifacts in RNA-Seq data. BMC Bioinformatics, 2016.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.