Use cases

Cluster your cells and find what makes each cluster different.

A 10x run counted, filtered, normalised, clustered and embedded, with marker genes for every cluster.

Join the waitlist
scrna.flow5 steps
transcriptomeINPUTFASTQCOUNTCellRangerQCScanpyEMBEDUMAPREFGRCh38

What is Single-cell analysis?

Single-cell RNA-seq analysis measures gene expression in thousands of individual cells, then groups cells with similar expression into clusters that usually correspond to cell types or states. A standard 10x Genomics workflow counts reads per cell with Cell Ranger, filters, normalises and clusters with Scanpy, embeds the cells with UMAP, and finds the marker genes that define each cluster.

Where bulk RNA-seq averages over every cell in a sample, single-cell RNA-seq keeps each cell separate. That shows which cell types are present, how their proportions shift with treatment, and which genes distinguish them, at the cost of sparser, noisier data for each cell.

Counting comes first: droplet methods tag every transcript with a cell barcode and a unique molecular identifier, and Cell Ranger uses them to build a matrix of genes by cells. Quality control then removes barcodes that are empty droplets, cells that were dying (a high share of mitochondrial reads) and doublets, where two cells shared one droplet. Normalisation puts the cells on a common scale, and the most variable genes carry the signal into a reduced representation.

Clustering connects each cell to its nearest neighbours and divides the resulting graph into communities, commonly with the Leiden algorithm. UMAP lays the cells out in two dimensions so the clusters can be seen, though distances on a UMAP plot should not be read literally. Marker genes, found by testing each cluster against the rest, are how clusters get their names.

What happens to your data

For groups profiling tissues, organoids or differentiation time courses.

Questions it answers

  • Which cell types are in this sample?
  • What marks each cluster?
  • Does the treatment shift the proportions?
  1. 1

    Cell Ranger

    Counts reads per gene per cell from the raw 10x data.

    Look at: Estimated cell number and median genes per cell.

  2. 2

    Scanpy

    Filters out empty droplets and dying cells, normalises and clusters.

    Look at: The filtering thresholds. Too strict and a real population disappears.

  3. 3

    UMAP

    Lays the cells out in two dimensions so similar cells sit together.

    Look at: Whether clusters split by sample or batch rather than biology.

What you take to lab meeting

Follow the protocol step by step

  • A UMAP coloured by cluster, sample or any gene

  • Marker genes for every cluster

  • The filtered count matrix

Common questions

How many cells do I need for single-cell RNA-seq?
Enough to capture the rarest population you care about in useful numbers. A population that makes up a small share of the sample needs proportionally more cells before it appears as a cluster at all, so plan around the biology rather than a fixed number.
What mitochondrial cut-off should I use?
There is no universal value. A common starting range is 10% to 20% of reads, but tissues with high energy demand, such as heart and muscle, sit naturally higher. Look at the distribution and cut at its tail rather than applying a fixed threshold (Heumos et al., 2023).
How should I read a UMAP plot?
As a map of similarity: cells that sit near each other have similar expression (McInnes et al., 2018). Clusters that sit apart are distinct, but the distances between clusters and their sizes on the plot are not reliable measures of how different they are.
Why do my clusters follow samples instead of cell types?
That is usually a batch effect: technical differences between samples outweighing the biology. Integrating the samples before clustering, with the batch as a covariate, is the standard correction. Check known marker genes afterwards to make sure the integration did not remove real differences.
Can I run single-cell analysis on a laptop?
For a typical experiment of tens of thousands of cells, the steps after counting run comfortably on a laptop. Counting the raw reads is the heaviest step and benefits from more memory and cores.

Sources

The papers behind the tools in this pipeline, and the advice on this page. Checked 25 October 2026.

  1. Zheng GXY, et al. (2017). Massively parallel digital transcriptional profiling of single cells. Nature Communications 8, 14049.
  2. Wolf FA, Angerer P, Theis FJ (2018). SCANPY: large-scale single-cell gene expression data analysis. Genome Biology 19, 15.
  3. Traag VA, Waltman L, van Eck NJ (2019). From Louvain to Leiden: guaranteeing well-connected communities. Scientific Reports 9, 5233.
  4. McInnes L, Healy J, Melville J (2018). UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. arXiv preprint 1802.03426.
  5. Heumos L, et al. (2023). Best practices for single-cell analysis across modalities. Nature Reviews Genetics 24, 550–572.

Read before you run it

A frank look at which single-cell analyses run comfortably on one machine, where the limits are, and what we are doing about the rest.

Single-cell on a laptop: what fits, and what does not yet

Field notes, 30 September 2026

Other starter pipelines

Run Single-cell on your own data.

OpenBionode is in early access. Tell us what you analyse and we’ll make sure the steps you need are ready when your seat opens.

Early access for research labs. No spam, just one note when your seat opens.