Lesson 9 of 20
Volcano plots — size across, confidence up
Fig. 3B, redrawn from the real GSE285342
You need 11,751 genes on one screen. For each of them you know two things —
- how much it changed — the log2 fold change. 1 means a doubling
- how much you believe it — the p-value. Smaller means less like chance
Put those on the two axes and a volcano appears: wide and low in the middle, sharp and high at both edges.
The data here is the paper's own — GSE285342, mouse LLC1 cells, 4 Ddx54-normal animals against 3 knockdowns.
Ddx54 itself went down — which is what a knockdown means. Drosha and Dicer1 went up; Myc, Cd38 and Cd47 went down. Everything the paper argued across six figures is in these seven lines.
Python
Runs in your browser
The same shape as Fig. 3B of the paper. The paper plotted microRNAs and we plotted mRNAs — the grammar is identical.
Python
Runs in your browser
Drawing a volcano means choosing two thresholds.
| Threshold | Common value | What it filters out |
|---|---|---|
| horizontal (|log2FC|) | 1 (= 2×) | changes too small to matter |
| vertical (p) | 0.05, or FDR 0.05 | changes that might be chance |
The two do different jobs. The horizontal one is biology, the vertical one is statistics. A gene that changed 1.02-fold is useless however small its p, and a tenfold change seen in one replicate is not believable.
🤔 Check yourself
Why is the y-axis of a volcano plot −log10(p)?
Answer p = 1e-6 and p = 1e-3 differ a thousandfold, yet on a linear axis both sit near zero. Take −log10 and they become 6 and 3 — a readable distance.
🤔 Check yourself
A threshold of |log2FC| ≥ 1 means what fold change?
Answer On log2, 1 is twofold, 2 is fourfold, 3 is eightfold. A log2FC of −1 means it halved.
🤔 Check yourself
A gene has a tiny p but a log2FC of 0.05. How should you read it?
Answer p answers whether there is a change; log2FC answers how much. With enough samples even a 3.5% change becomes significant. Hence two thresholds.