Expression and single cell - counts to clusters, decisions named
Lesson 4 of 6 · Your instrument, end to end
1 · Learn the move · Constraints + negative instructions
PARALLEL TRACK - complete only if pl1 pointed you here. Expression analysis, bulk or single cell, is a corridor of defaults: mitochondrial cutoff, minimum-genes threshold, clustering resolution - each shipped by a tool, none chosen by you, every one a scientific decision that moves your cell counts and clusters. The constraint discipline: the model may propose a threshold; every proposal is tagged DECISION with you as owner; nothing applies silently. The same discipline covers the layer question - raw, normalized, or scaled matrix, named at every step, because results from the wrong layer look fine - and the identifier question, gene symbols versus Ensembl IDs, symbol mapping being lossy and version-stamped. In qPCR, the same trap: delta-delta-Ct needs a declared reference gene and calibrator; a model handed a bare Cq table will invent both. It must not.
Author my expression pipeline from [10x Cell Ranger output / count matrix / qPCR export] under decision constraints: (1) every threshold - QC cutoffs, normalization choice, resolution - appears as DECISION: [value] - PROPOSED, applied only after I confirm; (2) every matrix operation names its layer (raw/normalized/scaled) in code and output; (3) gene identifiers: state symbol or Ensembl, and the annotation version; (4) nothing is dropped silently - every filter prints cells/genes before and after; (5) cluster outputs are described structurally, never labeled with cell types unless I provide the markers. [NEEDS: ...] for anything my description leaves open.
2 · Your turn. You write the prompt
Cell Ranger just finished on two treatment conditions and the filtered matrices are sitting in outs/. The last person who touched scanpy in your lab copied a tutorial's thresholds wholesale and nobody could defend them in lab meeting. Build the counts-to-clusters pipeline where every threshold has an owner and every layer is named.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.