Heatmap

A heatmap visualizes a matrix of values with color encoding intensity, typically with hierarchical clustering applied to rows and/or columns to reveal co-expression or similarity patterns. Four critical parameters determine the heatmap's appearance —and its scientific meaning: (1) the color scale, (2) the data scaling, (3) the distance metric, and (4) the linkage method. All four must be disclosed.

STEP 1 —When to Use

Use a heatmap when: you want to show intensity patterns across many features (genes, proteins, metabolites) and samples simultaneously; you want to use hierarchical clustering to reveal sample or feature groups without a priori assignment; you are showing normalized expression data where relative (not absolute) values are the message.

Heatmap color scale warning: rainbow color maps (jet, spectrum) are perceptually non-uniform, create false discontinuities, and are inaccessible to colorblind readers. Licklider defaults to perceptually uniform sequential (viridis, magma) or diverging (RdBu, BrBG) color scales.

[Image placeholder: Two heatmaps of the same data. Left: rainbow color map —a sharp visual transition appears at the middle of the scale that does not correspond to a data boundary. Right: viridis color map —smooth perceptual transition, correct representation. Same data, different visual impression.]

Rainbow color map (left) creates a false visual discontinuity that does not correspond to a data boundary. Perceptually uniform color map (right) correctly represents the data gradient.

STEP 2 —Journal Requirement

Color scale: type (sequential / diverging), range limits (minimum, midpoint, maximum), and color map name must be stated.
Scaling: row-scaled, column-scaled, or no scaling —and the direction and method (z-score, min-max) —must be stated.
Clustering: distance metric and linkage method for both rows and columns must be stated.
n (samples per group, if applicable) and feature count must be stated.
Clustering applied to: rows only, columns only, or both —must be stated.

STEP 3 —Licklider Implementation

Input

Numeric matrix (features xsamples)
Optional: sample annotation (grouping variable) for column color bars
Scaling: z-score by row (default for expression data), z-score by column, min-max, or none
Distance: Euclidean (default), Manhattan, 1-Pearson correlation, Spearman
Linkage: complete (default), average (UPGMA), Ward.D2, single
Color map: viridis (sequential default), RdBu (diverging default for z-scored data)

Output

Heatmap matrix with row and column dendrograms
Color scale bar labeled with scale range and units
Sample annotation color bars (group, batch, treatment) along the column axis
Feature significance stars optional (if statistical tests accompany the heatmap)
All parameters auto-logged in legend text and metadata

[Image placeholder: Licklider heatmap (50 features x24 samples). RdBu diverging color scale (z-scored by row, range ~ to +2). Complete linkage, Euclidean distance for rows and columns. Sample annotation bar at top (3 groups: control, treatment A, treatment B). Legend: "Color: z-score (row), complete linkage, Euclidean distance." Group clustering clearly visible.]

Licklider heatmap: perceptually uniform diverging color scale, z-scored by row, complete linkage / Euclidean distance, sample annotation bar, all parameters in legend.

STEP 4 —Draft Output (Draft / Needs review)

Heatmap of [N] features across [M] samples. Values are z-scored by row
(mean subtracted, divided by SD). Color scale: RdBu diverging palette
(blue: z 竕､ ~; white: z = 0; red: z >= +2). Hierarchical clustering was
applied to both rows (features) and columns (samples) using Euclidean
distance and complete linkage. Sample groups are indicated by the
annotation bar above the heatmap (n = 8 per group).