Coding agents helped make Seurat's core computations for single-cell data up to ten times faster

The Seurat developers have released a preview version of their package for single-cell RNA analysis

A preview of Seurat 5.6 became available on August 24. It speeds up the key steps that turn single-cell RNA measurements into a map of cell types and states.

Single-cell sequencing shows which RNAs are present in each cell. Biologists use these data to distinguish cell types and states. Comparing cells requires several steps: placing the measurements on a common scale, identifying genes that differ between cells, grouping similar cells, and finding markers for each group, which are genes that help determine its type.

The Seurat 5.6 preview updates these specific steps: data normalization, identification of variable and marker genes, preparation of data for comparison, construction of a compact map, identification of neighboring cells, and cell clustering. The branch can be installed directly from GitHub.

The developers attribute most of the performance gain to moving the main computations to C++, a programming language suited to fast numerical processing. Normalization, compact map construction, and clustering can distribute work across multiple processor cores, with the user setting the number of threads.

In June, the laboratory reported that moving parts of Seurat from R to C++ made some analysis steps ten times faster or more. The current beta branch gives users access to the rewritten analysis steps and control over multithreading.