Skip to main content

Volcano plot

A volcano plot puts log₂ fold change on the x axis against −log₁₀(p-value) on the y axis — so the points that matter are the ones high up and far out. Points passing both cutoffs are coloured up (right) or down (left), everything else is grey, and the threshold lines are drawn for you.

SELECT kuva_render(to_json({
'title': 'Tumour vs normal',
'x_axis': {'name': 'log2 fold change'},
'y_axis': {'name': '-log10(p-value)'},
'series': [{
'type': 'volcano',
'points': pts,
'legend': 'status'
}]
})) AS chart
FROM (
SELECT list({'name': gene, 'log2fc': log2fc, 'pvalue': pvalue}) AS pts
FROM read_csv_auto('{{DFK_BASE_URL}}data/volcano.tsv')
);

Two thresholds do all the classification: fc_cutoff (default 1, i.e. a two-fold change) and p_cutoff (default 0.05). Dashed lines appear at ±fc_cutoff and at −log10(p_cutoff).

Gene labels​

label_top labels the n most significant points, which is how the handful of genes worth naming get named without labelling 20 000 of them. Four placement styles are available:

label_stylePlacement
"nudge"Labels sorted by x and nudged vertically to reduce stacking (default)
"exact"At the point, with no adjustment — for sparse data or post-processing
{"offset_x": 14, "offset_y": 16}Offset by pixels, with a short leader line back to the point
"repel"Force-directed (ggrepel / adjustText style): labels pushed off each other and off the points, with thin leader lines
SELECT kuva_render(to_json({
'title': 'Top hits labelled',
'x_axis': {'name': 'log2 fold change'},
'y_axis': {'name': '-log10(p-value)'},
'series': [{
'type': 'volcano',
'points': pts,
'label_top': 12,
'legend': 'status'
}]
})) AS chart
FROM (
SELECT list({'name': gene, 'log2fc': log2fc, 'pvalue': pvalue}) AS pts
FROM read_csv_auto('{{DFK_BASE_URL}}data/volcano.tsv')
);

The arrow style is the one to use when the labels would crowd the high-significance corner, since the leader line lets a label sit at a comfortable distance from its point:

SELECT kuva_render(to_json({
'title': 'Arrow labels',
'x_axis': {'name': 'log2 fold change'},
'y_axis': {'name': '-log10(p-value)'},
'series': [{
'type': 'volcano',
'points': pts,
'label_top': 10,
'label_style': {'offset_x': 14, 'offset_y': 16},
'legend': 'status'
}]
})) AS chart
FROM (
SELECT list({'name': gene, 'log2fc': log2fc, 'pvalue': pvalue}) AS pts
FROM read_csv_auto('{{DFK_BASE_URL}}data/volcano_genes.tsv')
);

Adding the nudges' vertical spread is derived from the data, so the same label_top` produces a tighter or looser cloud depending on how clustered the top hits are.

Thresholds​

Stricter cutoffs move the lines inward on the x axis and upward on the y axis, shrinking the coloured corners. They are the first thing to check when two volcano plots of the same comparison disagree.

SELECT kuva_render(to_json({
'title': 'Stricter cutoffs',
'x_axis': {'name': 'log2 fold change'},
'y_axis': {'name': '-log10(p-value)'},
'series': [{
'type': 'volcano',
'points': pts,
'fc_cutoff': 2,
'p_cutoff': 0.01,
'color_up': 'darkorange',
'color_down': 'mediumpurple',
'label_top': 8,
'legend': 'status'
}]
})) AS chart
FROM (
SELECT list({'name': gene, 'log2fc': log2fc, 'pvalue': pvalue}) AS pts
FROM read_csv_auto('{{DFK_BASE_URL}}data/volcano.tsv')
);

Colours​

FieldDefaultUsed for
color_upfirebricklog2fc ≥ fc_cutoff and p ≤ p_cutoff
color_downsteelbluelog2fc ≤ −fc_cutoff and p ≤ p_cutoff
color_ns#aaaaaaEverything else
SELECT kuva_render(to_json({
'title': 'House colours',
'x_axis': {'name': 'log2 fold change'},
'y_axis': {'name': '-log10(p-value)'},
'series': [{
'type': 'volcano',
'points': pts,
'color_up': '#d62728',
'color_down': '#1f77b4',
'color_ns': '#cccccc',
'point_size': 3.5,
'legend': 'status'
}]
})) AS chart
FROM (
SELECT list({'name': gene, 'log2fc': log2fc, 'pvalue': pvalue}) AS pts
FROM read_csv_auto('{{DFK_BASE_URL}}data/volcano.tsv')
);

Zero and extreme p-values​

A p-value of exactly 0 cannot be log-transformed. Such points are clamped to the smallest non-zero p-value in the data — which means the y axis tops out wherever the most significant gene happens to be, and two volcano plots of different datasets end up on different scales.

pvalue_floor sets that ceiling explicitly, so several plots can share a y axis:

SELECT kuva_render(to_json({
'title': 'Pinned y ceiling',
'x_axis': {'name': 'log2 fold change'},
'y_axis': {'name': '-log10(p-value)'},
'series': [{
'type': 'volcano',
'points': pts,
'pvalue_floor': 1e-10,
'legend': 'status'
}]
})) AS chart
FROM (
SELECT list({'name': gene, 'log2fc': log2fc, 'pvalue': pvalue}) AS pts
FROM read_csv_auto('{{DFK_BASE_URL}}data/volcano.tsv')
);

Fields​

FieldTypeWhat it sets
pointspoint[]Required. Each entry is {name, log2fc, pvalue}.
fc_cutoffnumberThe |log2FC| threshold (default 1).
p_cutoffnumberThe p-value threshold (default 0.05).
color_up / color_down / color_nsstringThe three classifications' colours.
point_sizenumberCircle radius in pixels (default 3).
label_topintegerLabel the n most significant points (0 = none, the default).
label_stylestring | object"nudge" (default) · "exact" · "repel" · {offset_x, offset_y}.
pvalue_floornumberExplicit p-value floor for the −log10 transform.
legendstringAny non-empty value turns the up / down / NS legend on.

Notes​

  • points must not be empty, and every point needs name, log2fc and pvalue.
  • pvalue is the raw p-value, not −log10 of it.
  • p-values above 1 or at or below 0 are nonsensical; the floor handles the second case, the first is left to draw at a negative y rather than being silently dropped.
  • label_top labels by significance, not by fold change — the labels go to the tallest points, which is usually but not always what you want to show.
  • fc_cutoff is compared against the absolute value, so the two dashed x lines are symmetric.

See also​

  • kuva — Volcano plot — the plotting library's own reference for this chart.
  • Manhattan — significance by genomic position instead of effect size.
  • Q-Q plot — check the p-value distribution itself before trusting the peaks.