Assay

Single-Cell RNA-seq: sanity checks and pitfalls

Droplet-based capture (10x Chromium is the default) yields a sparse cell-by-gene UMI matrix that goes through cell QC, normalization, feature selection, PCA, neighbor graph, clustering and 2-D embedding. Every one of those steps has a knob that changes the biology you report: mitochondrial and count thresholds remove real cell types, resolution invents clusters, and integration can erase the condition effect you were funded to find. Doublets, ambient RNA and sample-confounded batches are present in essentially every dataset and are only a problem when nobody checks.

Who this is for: Immunologists, cancer biologists and core-facility analysts who have Cell Ranger output for 4-20 samples and need clusters they can defend to a reviewer. Most are working in Seurat or Scanpy and choosing thresholds, resolutions and integration methods by copying a vignette.