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. Assume cells A and B have the same RNA composition and differ only in capture. A gene has 20 of A's 200 counts and 40 of B's 400 counts. Scale each cell to 1,000. What are the gene's scaled counts?

    • 100 for A and 100 for B
    • 100 for A and 200 for B
    • 200 for A and 100 for B

  2. In the common workflow taught here, which representation is used to build the neighbour graph?

    • UMAP coordinates from the displayed cell layout
    • PCA component scores from selected gene features
    • Normalised counts for all genes before selecting features

  3. A zero in a single-cell RNA count matrix proves that no RNA from that gene was present in the cell.

    Circle one:   True   False

  4. Two clusters appear far apart in a two-dimensional UMAP view. Which conclusion is supported by that view alone?

    • Separation is visible in this layout; the underlying cell identities remain unresolved.
    • The large gap establishes that the clusters contain different cell types without needing other evidence.
    • The gap provides a calibrated measure of biological difference, with larger gaps indicating greater differences.

  5. Normalised values in three cells are A: 6, 6, 6 and B: 2, 6, 10. Expected technical noise is similar for these genes. Which is the better candidate for a variable-feature set?

    • Gene A, because stable values suggest less technical noise.
    • Both genes, because their matching average values suggest equal value for grouping.
    • Gene B, because its values change across cells despite a matching mean.

  6. Increasing clustering resolution splits a marker-supported cluster into two groups. What is the best next step?

    • Use finer cell-type labels, then check whether marker differences support the new names.
    • Compare markers and sample effects, retaining a broad label while its evidence holds.
    • Set the broad label aside, then use the split to guide a fresh identity assignment.

  7. Total-count scaling removes every doublet, ambient-RNA signal and batch effect from a UMI count matrix.

    Circle one:   True   False

  8. In a teaching example, a tissue- and species-matched reference links markers P and Q to type T. A cluster enriches both across donors, and staining agrees. Which label is best supported?

    • A distinct state within type T, supported by the repeated marker pattern and agreeing staining
    • Type-T-like RNA profiles, with protein identity left unsupported
    • Type-T-like cells, with marker and staining support recorded alongside uncertainty
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. 100 for A and 100 for B · 20 / 200 × 1,000 = 100 and 40 / 400 × 1,000 = 100. Each cell assigns one tenth of its counts to the gene.
  2. PCA component scores from selected gene features · The workflow links cells with similar PCA scores. Clustering uses that graph; it need not use the two displayed UMAP coordinates.
  3. False · A zero records no counted molecule. RNA may have been absent, or present RNA may have escaped detection.
  4. Separation is visible in this layout; the underlying cell identities remain unresolved. · UMAP separation is a layout observation. Its gaps are not calibrated biological distances or sufficient evidence of different cell types.
  5. Gene B, because its values change across cells despite a matching mean. · Both means are 6, but A is constant and B varies. With similar expected noise, B is the better candidate for further feature-selection checks.
  6. Compare markers and sample effects, retaining a broad label while its evidence holds. · Higher resolution can create finer partitions. Investigate their biological and technical explanations while retaining any broad identity that remains supported.
  7. False · Scaling changes the count scale. It does not identify doublets, determine molecule origins or remove every processing-related effect.
  8. Type-T-like cells, with marker and staining support recorded alongside uncertainty · The matched reference, repeated markers and agreeing stain support a provisional T-like identity. The stain adds independent support; the prompt supplies no evidence distinguishing a particular state within type T.
Worksheet · LightMySky