Zubair Khalid

Virologist/Molecular Biologist | Veterinarian | Bioinformatician

Conventional & Molecular Virology • Vaccine Development • Computational Biology

Dr. Zubair Khalid is a veterinarian and virologist specializing in conventional and molecular virology, vaccine development, and computational biology. Dedicated to advancing animal health through innovative research and multi-omics approaches.

Dr. Zubair Khalid - Veterinarian, Virologist, and Vaccine Development Researcher specializing in Computational Biology, Multi-omics, Animal Health, and Infectious Disease Research

Blog · Guides · Published 2026-07-12

Single-Cell RNA-seq Workflow: A Practical Analysis Roadmap

This guide maps the essential decision points in a single cell RNA sequencing (scRNA seq) analysis from a count matrix to biological interpretation. It is intended for researchers who have generated or acquired scRNA seq data and need to make informed, reproducible choices at each stage. The workflow described relies on established tools and publicly available resources. Automation is limited, many decisions require domain knowledge about the tissue, protocol, and biological question.

At a Glance

Stage Key Decisions
Count matrix generation Mapping tool, gene vs. transcript quantification, ambient RNA handling
Quality control Cell filtering thresholds per gene count, mitochondrial percentage, library size
Normalization Library size scaling, log transform, SCTransform, or alternative methods
Dimensionality reduction Number of principal components, use of harmony, scVI, or other embedding corrections
Clustering Algorithm (Louvain, Leiden), resolution parameter, stability validation
Cell type annotation Manual marker inspection, automated label transfer, cluster tree analysis
Differential expression Pseudobulk vs. per cell, model selection, effect size
Trajectory and downstream Graph based pseudotime, RNA velocity, gene module detection

Decision Criteria That Cannot Be Automated

No automated pipeline can replace careful, context aware choices. The most critical decisions include:

Quality thresholds. The proportion of mitochondrial reads considered high depends on tissue type and cell viability. For example, stressed or damaged cells in a pancreatic tumor sample may show higher mitochondrial content than healthy immune cells. You must inspect distributions and set cutoffs per dataset (see Galaxy Training Network for practical tutorials on scRNA seq QC).

Normalization strategy. Log transformation followed by scaling is standard but can distort rare cell types. Methods like SCTransform (from Bioconductor) model technical noise more robustly. The choice should match downstream analysis goals.

Batch correction. If samples were processed in separate runs, you must decide whether to correct for batch effects and which algorithm to use. Overcorrection can remove biological variation.

Clustering resolution. A low resolution yields broad clusters, high resolution splits fine subpopulations. There is no single correct value. Use cluster stability metrics and biological plausibility to guide the choice.

Cell type annotation. Automated annotation using reference datasets is fast but may mislabel novel or tissue specific populations. Manual inspection of known markers is essential.

Practical Workflow

1. Generate the count matrix

Raw sequencing data (FASTQ) can be obtained from public repositories such as the NCBI Sequence Read Archive. After mapping and quantification using tools like Cell Ranger, STARsolo, or kallisto bus, you obtain a gene cell count matrix. At this stage, consider ambient RNA removal (e.g., with SoupX) which is critical for tissues with high background lysis.

2. Quality control and filtering

Remove cells with low library size, fewer than 200 500 detected genes, and high mitochondrial gene fraction (typically above 10 20% for most tissues, but check your data). Doublet detection (using Scrublet or DoubletFinder) should follow. Retain only high quality cells.

3. Normalization and feature selection

Normalize the count matrix to account for differences in sequencing depth. A common approach in Seurat is NormalizeData (log transform with scale factor) followed by FindVariableFeatures. For datasets with strong technical noise, consider more robust options. EMBL EBI Training provides step by step exercises on normalization strategies.

4. Dimensionality reduction and batch correction

Perform PCA on highly variable genes. Inspect the elbow plot to choose a meaningful number of principal components (typically 15 50). If batch effects are present, apply integration methods such as Harmony, Seurat CCA, or scVI. Verify that integration does not collapse distinct cell types by evaluating batch mixing versus biological separation.

5. Clustering

Use a graph based algorithm (Louvain or Leiden) on the reduced embedding. Iterate over a range of resolution parameters and use metrics like silhouette score or the adjusted Rand index on subsamples to assess stability. Visualize clusters on a UMAP or t SNE embedding, but remember that these visualizations are not quantitative and can be misleading.

6. Cell type annotation

Identify cluster markers using differential expression (e.g., FindAllMarkers). Cross reference with known marker genes from literature or databases such as GIDISdb (see PubMed 42423370) for immune cell signatures. For tissues with limited prior annotation, automatic label transfer from reference atlases (e.g., Human Cell Atlas) can provide initial labels. Always validate with independent markers.

7. Differential expression between conditions

If comparing groups (e.g., disease vs. control), treat clusters as pseudobulk aggregates to avoid inflated p values due to multiple cells per sample. Use methods like DESeq2 or edgeR on summed counts per sample per cluster. This approach accounts for biological variability between samples more accurately than treating each cell as independent.

8. Trajectory and downstream analyses

For developmental processes or transitions, infer pseudotime using Slingshot or Monocle 3. RNA velocity (via scVelo) can give directionality. Additionally, analyze gene regulatory networks or ligand receptor interactions. Methods like VDJ REMIX (PubMed 42412835) can uncover repertoire modules in immune cells.

Common Mistakes

Over filtering or under filtering. Setting mitochondrial threshold too low can remove viable cell types. Setting it too high can retain damaged cells. Always assess the distribution of metrics across clusters.

Ignoring ambient RNA. Protocols with high background (e.g., droplet based 10x) may include counts from lysed cells or free RNA. Not correcting can create false clusters of low quality cells.

Using raw counts for differential expression without pseudobulk. Comparing each cell individually without accounting for sample level replication leads to many false positives.

Relying solely on clustering for biological discovery. Clusters are computational constructs. Always validate findings with independent assays like flow cytometry or spatial transcriptomics.

Failing to document software versions and parameter changes. Reproducibility requires a clear record.

Limits and Uncertainty

Single cell RNA seq captures transcript abundance at low coverage per cell. Dropout events (zeros for expressed genes) are common and limit detection of lowly expressed transcripts. Sequencing depth affects sensitivity. The technology cannot measure protein abundance or post translational modifications. Batch effects can be strong and are never fully removed. Interpretation beyond correlation requires validation. For woody plants and other non model organisms, reference genome quality may be poor, affecting alignment and quantification (PubMed 42409600 highlights challenges in forest trees). Finally, cell type nomenclature is not standardized across tissues, leading to inconsistent annotation.

Frequently Asked Questions

What is a practical minimum number of cells per cluster for reliable analysis?
At least 50 cells per cluster is recommended for robust marker detection and differential expression. Rare populations with fewer cells can be analyzed but interpretation requires caution and additional validation.

How do I choose between 10x Genomics and Smart seq2 for my study?
10x captures many cells at lower coverage (suitable for population mapping and rare cell discovery). Smart seq2 offers full length transcript coverage with higher per cell depth (better for isoform and allele specific expression). Consider your budget and biological question.

Can I reuse public scRNA seq data for my own analysis?
Yes, provided you download raw count matrices or FASTQ files from repositories like the NCBI Sequence Read Archive. However, metadata quality varies. Always check sample level covariates and processing protocols before re analysis.

How do I validate novel cell type predictions from my data?
Use independent methods: immunohistochemistry, flow cytometry with specific antibodies, or spatial transcriptomics to confirm marker co expression. Also, compare to multiple reference datasets. For example, automatic extraction of mesenchymal stromal cells from dental pulp data (PubMed 42415183) demonstrates validation through iterative refinement.

References and Further Reading

  1. NCBI Bookshelf. Overview of single cell RNA sequencing methods. Available at: https://www.ncbi.nlm.nih.gov/books/
  2. EMBL EBI Training. Single cell RNA seq analysis tutorials. Available at: https://www.ebi.ac.uk/training/
  3. Galaxy Training Network. Single cell RNA sequencing workflow. Available at: https://training.galaxyproject.org/
  4. Bioconductor. Packages for single cell analysis. Available at: https://bioconductor.org/
  5. NCBI Sequence Read Archive. Repository for raw sequencing data. Available at: https://www.ncbi.nlm.nih.gov/sra
  6. Single Cell RNA Sequencing of Pancreatic Circulating Tumor Cells. Methods Mol Biol. Available at: https://pubmed.ncbi.nlm.nih.gov/42426462/
  7. GIDISdb: a gene expression database for exploring human immune responses in infectious diseases. Database (Oxford). Available at: https://pubmed.ncbi.nlm.nih.gov/42423370/
  8. Protocol for minimized bias profiling of liver and visceral adipose tissue in mice using integrated single nucleus RNA sequencing. STAR Protoc. Available at: https://pubmed.ncbi.nlm.nih.gov/42418327/
  9. Automatic extraction of mesenchymal stromal cells from single cell RNA sequencing data of human dental pulp. BioData Min. Available at: https://pubmed.ncbi.nlm.nih.gov/42415183/
  10. VDJ REMIX: REpertoire Module Identification and eXploration. Bioinformatics. Available at: https://pubmed.ncbi.nlm.nih.gov/42412835/

Related Articles