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

Category: Guides

Single Cell Transcriptomics

Single cell transcriptomics is a set of techniques that measure gene expression in individual cells, revealing heterogeneity hidden in bulk RNA sequencing. This guide provides a source bounded, practical framework for anyone planning, executing, or interpreting a single cell transcriptomics experiment: bench scientists, bioinformaticians, and reviewers. You will learn core concepts, decision points, a step by step workflow, quality checks, common mistakes, and the limits of what these data can tell you.

According to authoritative training materials from the EMBL EBI Training, single cell RNA sequencing (scRNA seq) captures the transcriptome of thousands of cells in parallel, enabling discovery of rare cell types, developmental trajectories, and disease specific transcriptional programs. The raw data are housed in public repositories such as the NCBI Sequence Read Archive, where you can find thousands of published datasets.

At a Glance

Aspect Key Information
What it measures Gene expression at single cell resolution
Typical output Count matrix: genes x cells
Main technologies 10x Genomics, Drop seq, Smart seq, MARS seq
Key data size 1,000 to 100,000+ cells per experiment
Core analytic steps Pre processing, normalization, dimensionality reduction, clustering, differential expression, cell type annotation
Typical cost per sample $1,000 - $5,000 (library prep and sequencing)
Primary limitation Dropout zeros and technical noise

Core Concepts

Single cell transcriptomics starts with isolating individual cells, capturing their mRNA, reverse transcribing it into cDNA, and sequencing the library. The result is a digital count matrix where each row is a gene and each column is a cell. The key challenge is that any given gene may be undetected in a cell even when it is truly expressed (dropout events).

The field relies on two major families of protocols. Droplet based methods (e.g., 10x Genomics) encapsulate cells in oil droplets with barcoded beads, enabling massively parallel capture. Full length methods (e.g., Smart seq2) provide higher sensitivity per cell but lower throughput. A foundational resource for understanding these differences is the NCBI Bookshelf, which offers free textbooks on sequencing technologies.

Data analysis typically follows a standard pipeline. After raw reads are aligned to a reference genome, the count matrix undergoes filtering to remove empty droplets and low quality cells. Normalization accounts for differences in sequencing depth. Dimensionality reduction using PCA or UMAP produces a two dimensional view of cell populations. Clustering algorithms assign cells into groups, and differential expression analysis identifies marker genes. The Galaxy Training Network provides hands on tutorials for each of these steps.

Decision Points

Before starting an experiment, you must decide on several parameters. The choice of platform depends on your biological question and budget.

Throughput vs sensitivity. Droplet based methods yield thousands of cells per run but capture only the 3 prime end of transcripts, missing splice variants. Full length methods capture more of each transcript but at lower cell numbers. For rare cell types, higher sensitivity is critical. For mapping cell atlases, throughput wins.

Sample preparation. Single cell protocols require a viable single cell suspension. Tissues like brain or tumor must be dissociated without losing cell surface markers or inducing stress. Cryopreservation or direct processing from fresh tissue each have trade offs.

Sequencing depth. Deeper sequencing does not always improve accuracy. Most variability comes from biological heterogeneity, not sampling. Typical depth is 50,000 reads per cell. A practical guide from the Galaxy Training Network recommends focusing on cell number over read depth for most applications.

Multiplexing. To reduce batch effects, you can pool samples using cell hashing or genetic demultiplexing. This increases complexity but is essential when comparing conditions.

Practical Workflow

The workflow below represents a standard implementation using open source tools available through the Bioconductor project.

  1. Raw data processing. Demultiplex barcodes using software like Cell Ranger (10x) or zUMIs. Align reads to the reference genome with STAR or hisat2. Generate a count matrix.

  2. Quality control. Filter cells based on three metrics: number of unique genes detected (usually 500 to 5000), total UMI counts (e.g., > 1000), and mitochondrial fraction (typically < 20% to exclude dead or stressed cells). Automated QC is available in packages like scater.

  3. Normalization. The default approach is library size normalization scaled to 10,000 counts per cell, followed by log transformation. Alternative methods like scran use pooling to correct for composition biases. Normalization is discussed in depth in the EMBL EBI Training materials.

  4. Batch correction. If samples were processed in multiple batches, use methods like Harmony, Seurat CCA, or MNN. Visualize before and after to confirm removal of batch effects while preserving biological variation.

  5. Dimensionality reduction and clustering. Run PCA on the top 2000 highly variable genes, then compute UMAP. Cluster cells using graph based community detection (e.g., Louvain or Leiden). Choose resolution based on the granularity needed.

  6. Cell type annotation. Identify cluster markers with non parametric tests (Wilcoxon rank sum). Use known marker databases (e.g., PanglaoDB, CellMarker) and canonical literature. Manual annotation remains the gold standard.

  7. Downstream analysis. Depending on your question, you may perform trajectory inference (Monocle, slingshot), differential expression between conditions, or gene regulatory network inference.

Quality Checks

Throughout the workflow, you must verify that your data are reliable.

Doublet detection. Doublets are two cells captured in one droplet. Tools like DoubletFinder or scDblFinder estimate doublet rate (typically 1-10%) and flag likely doublets for removal.

Mitochondrial content. High mitochondrial reads indicate cell damage. Remove cells above a threshold agreed in the field (often 20% for droplet data, 10% for Smart seq).

Library complexity. A low number of genes per cell suggests poor capture or damaged cells. Conversely, a very high gene count may indicate doublets.

Integration quality. After batch correction, confirm that cell types from different batches mix appropriately. Use metrics like silhouette width to assess clustering stability.

Common Mistakes

  1. Ignoring batch effects. Running all cells together without correction can produce clusters that reflect technical variation rather than biology. Always check for batch patterns in your UMAP.

  2. Over clustering. Using too high resolution creates many small clusters that have no interpretable marker genes. Start with low resolution and validate robustness.

  3. Under filtering. Retaining low quality cells inflates noise and can create artificial clusters. Be stringent but not overly aggressive.

  4. Annotation by default. Automated annotation tools rarely replace manual curation. Always validate with literature and known markers.

  5. Misinterpreting dropout. A zero in the count matrix does not mean no expression. Imputation can be risky, modern analysis (e.g., with sctransform) handles dropout better.

Limits of Interpretation

Single cell transcriptomics has important constraints. Dropout leads to false negatives. The data are also compositional: expression values are relative within each cell, not absolute. Comparing expression across conditions must account for this.

Another limit is the multicellular origin of many biological processes. Gene expression in a single cell snapshot does not reveal intercellular interactions or spatial context, although emerging spatial transcriptomics methods now address that gap. Recent studies, such as those examining tumor microenvironment signatures in PubMed ID 42443752 and super enhancers in lung adenocarcinoma PubMed ID 42443661, illustrate how scRNA seq is combined with other omics to overcome some limitations.

Statistical power for rare populations is limited by cell number. Trajectory inference assumes a continuous process, but validation with pseudotime alone is insufficient. For clinical applications, findings require orthogonal validation (e.g., flow cytometry, imaging).

Frequently Asked Questions

Q: How many cells are needed for a robust single cell experiment? A. For characterizing major cell types, 1,000 5,000 cells per sample may suffice. For rare populations (less than 1%), you need 10,000 or more cells. The Galaxy Training Network suggests designing for the rarest cell type of interest.

Q: Can I use frozen tissue? A. Yes. Cryopreservation protocols exist for single cell RNA seq, but viability may drop. Droplet based methods tolerate moderate viability (70%+). Always check your protocol.

Q: How do I choose between 10x Genomics and Smart seq2? A. For large scale atlases and most discovery experiments, 10x is cost effective. For full length transcript coverage, splicing analysis, or very few cells, Smart seq2 remains superior. See comparisons in Bioconductor documentation.

Q: What is the best way to submit data? A. Submit raw FASTQ files to the NCBI Sequence Read Archive and processed count matrices to GEO or EMBL EBI ArrayExpress. Include a metadata table with sample conditions.

References and Further Reading

Related Articles