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

A common assumption is that single-cell RNA-seq needs a cluster. For the part most people spend their time on, it does not. Most single-cell experiments in an academic lab profile tens of thousands of cells, not millions, and at that size the analysis after counting runs comfortably on one recent machine.

What fits on a laptop

Once you have a matrix of genes by cells, the standard downstream workflow is light enough for a laptop:

• Quality control: removing empty droplets, dying cells and doublets, and choosing the thresholds by looking at the distributions.
• Normalisation and the selection of highly variable genes.
• Principal components, the nearest-neighbour graph and clustering.
• UMAP and t-SNE embeddings for looking at the clusters.
• Marker gene tests between clusters.

These are the steps Scanpy was built for, and it handles experiments of this size in memory on one machine.

What does not fit, yet

Generating the count matrix from raw reads is the heavy part. Cell Ranger, the standard tool for 10x Genomics data, aligns every read and assigns it to a cell barcode and a unique molecular identifier. It needs a large reference index held in memory and benefits from many cores, which is more than most laptops have to spare.

For now the practical route is to start from the count matrix your sequencing provider or core facility already produces. Most supply one alongside the raw reads, and the analysis that answers the biological question starts there.

What we are doing about the rest

The same pipeline will run on a cluster or in the cloud when that support arrives, without being rebuilt: the steps and settings stay the same, only where they run changes. Until then we would rather say plainly where the edge is than promise a laptop will do something it will not.

To follow the analysis step by step, from counting to marker genes, use the single-cell protocol, or see the single-cell use case.