The problem. Single-cell analysis has a dozen decision points before you ever see a UMAP, and most of them silently determine the result. This tutorial is the field’s shared answer to “what should the pipeline actually be,” and it underpins my AML scRNA-seq project, so I read it as a retro of my own choices.

The idea. It’s a stage-by-stage workflow with defensible defaults. Preprocessing: quality filtering (counts, genes, mitochondrial fraction), correcting ambient RNA and removing doublets, then normalization, highly-variable-gene selection, scaling, and dimensionality reduction — with the standing warning that how you filter changes everything downstream. Downstream: clustering, cluster annotation (the genuinely hard, human-judgment part), marker/differential expression, compositional analysis, and trajectory/pseudotime inference. It closes with a reproducible worked example so the advice is executable, not just prose.

Why it matters. The value is the audit. Reading it, the questions write themselves: did I filter too aggressively on mito percentage? Did I validate cluster annotations against markers or just trust the algorithm? Is my trajectory analysis reading real biology or over-interpreting noise (the exact caution the repressilator’s variability taught)? Turning those into an honest retrospective is worth more than any single new method.

Verdict. Still the best on-ramp, with the caveat that it predates the current wave — scVI-class integration and single-cell foundation models have matured since, and a modern reader should pair it with the 2023 best-practices update. But for the fundamentals of not fooling yourself, it holds. Next: a candid “what I’d change in my AML workflow with hindsight.”