Single-Cell Analysis: Normalisation, Embedding and Naming Cell States · seed 1 · A4, ink-friendly. The answer key prints on its own page for grown-ups.

From soup of reads to named cell states

Science · Genetics & Evolution · ages 23-24
Name ______________________   Date ____________

  1. Why do raw counts need scaling before you compare two cells?

    • Because capture and amplification differ for technical reasons
    • Because every cell holds the same total RNA
    • Because barcodes fade over time

  2. What turns a soup of single-cell reads into per-cell data?

    • Deeper sequencing of the pool
    • The barcode tag carried by each fragment
    • Staining the cells afterwards

  3. A zero in the count table proves the gene was switched off in that cell.

    Circle one:   True   False

  4. Two clusters sit far apart on the two-dimensional plot. What may you claim?

    • Only that the method placed them apart; distances there are not measurements
    • That they must be different cell types
    • The plot distance measures their true biological distance

  5. What does feature selection keep, and why?

    • The most highly expressed genes, since they are safest
    • One marker per known type, to save compute
    • The genes varying most across cells, so states stand out

  6. You rerun the analysis with different clustering settings and a group splits in two. What follows?

    • Keep the original labels unchanged
    • Treat conclusions resting on that group as fragile until tested further
    • Merge everything into one cluster

  7. A clustering that invents false splits from dropout zeros is acceptable if the plot looks clean.

    Circle one:   True   False

  8. Which cluster label would you trust?

    • A name borrowed from a bulk study of another tissue
    • A name from one plot with no markers listed
    • A name backed by sensible markers that repeat and pass staining
LightMySky · lightmysky.comW1-mt_diYHLaLmij-s1

Answer key

For grown-ups. Fold this page away before handing over the rest.

From soup of reads to named cell states W1-mt_diYHLaLmij-s1

  1. Because capture and amplification differ for technical reasons · Uneven capture makes depth differ, so unscaled counts mislead.
  2. The barcode tag carried by each fragment · Each tag reveals which cell its fragment came from.
  3. False · Dropout hides real messages, so zeros mean unknown, not absent.
  4. Only that the method placed them apart; distances there are not measurements · The picture suggests groups but its distances are not data.
  5. The genes varying most across cells, so states stand out · Flat genes add static, while varying genes carry the state signal.
  6. Treat conclusions resting on that group as fragile until tested further · Only conclusions that survive changed settings deserve trust.
  7. False · Methods must respect the zero gap; pretty plots never excuse false states.
  8. A name backed by sensible markers that repeat and pass staining · Markers that repeat and survive a second test earn the label.
Worksheet · LightMySky