The impact of package selection and versioning on single-cell RNA-seq analysis

Standard single-cell RNA sequencing (scRNA-seq) workflows convert raw reads into cell-gene count matrices and then perform filtering, highly variable gene selection, dimensionality reduction, clustering, and differential expression analysis. Seurat and Scanpy are the most widely used implementations of these workflows and are generally assumed to yield similar results. Here, we examine the underlying algorithms of both packages and show that their outputs differ substantially across multiple analytic steps. The magnitude of Seurat-Scanpy differences is comparable to the variability introduced by sequencing fewer than 5% of reads or analyzing fewer than 20% of cells. We further find that software version changes can alter results, particularly in differential expression. Our findings underscore the need for users to critically evaluate analysis tools and for developers to prioritize transparency and reproducibility.