Lesson 20 of 20
Wrapping up — before you send a figure out
what you can say, and what you cannot
Twenty lessons behind us. Eight figures, 62 panels, nineteen kinds — taken apart one by one and redrawn.
Squeezed onto a single page, this is what you learned.
| Question | Answer | Lesson |
|---|---|---|
| Which chart should I use? | the kind of data first, then the question | 3 |
| Why so many bars? | because people read length best | 2 |
| What about the axis? | bars start at zero. Log it if it moves by multiples | 5 |
| What about colour? | pick one of sequential / diverging / categorical and do not mix | 6 |
| How do I show a distribution? | if n is small, show every point | 8 |
| And the error bars? | always say whether they are SD, SEM or CI | 11 |
| UMAPs and networks? | position means nothing. Let another chart carry the claim | 13 · 17 |
| How is a figure built? | wide → shortlist → pinpoint | 19 |
What you can say
Claims the figures support
· “Cd47 went down in the knockdown” — all seven replicates are visible and do not overlap
· “CD8A is higher in the inflamed group” — the two violins barely overlap
· “14 of this paper's 62 panels are bars” — we counted
· “CD8A is higher in the inflamed group” — the two violins barely overlap
· “14 of this paper's 62 panels are bars” — we counted
Claims they do not support
· “Far apart on the UMAP, so very different cells” — distance is not preserved
· “Blocks in the heatmap, so there are two groups” — noise makes blocks too
· “Central in the network, so a central gene” — position comes from the seed
· “The bar is twice as tall, so it is twice as much” — only if the axis starts at zero
· “Blocks in the heatmap, so there are two groups” — noise makes blocks too
· “Central in the network, so a central gene” — position comes from the seed
· “The bar is twice as tall, so it is twice as much” — only if the axis starts at zero
Read that list again. The names that were unfamiliar in lesson 1 are now names whose job you know.
Python
Runs in your browser
🤔 Check yourself
Why must a representative image be accompanied by quantification?
Answer Bias enters at the moment of choosing. Only the quantification lets the reader judge whether that one picture is the rule or the exception.
🤔 Check yourself
If this course repeated one rule from beginning to end, which was it?
Answer A bar hides the distribution, a UMAP hides distance, a box hides the modes, a z-score hides absolute level. Knowing what you hid is the whole of reading and drawing figures.
🤔 Check yourself
Where should you look first at an unfamiliar chart?
Answer The axis is the quietest part of a chart and the part that changes the conclusion most. Remember lesson 18, where a truncated axis turned 7% into a tripling.