Heatmap

A heatmap displays values from multiple numeric columns as a color-coded grid. Each row in the dataset becomes a row in the figure; each selected numeric column becomes a column. The color of each cell represents the magnitude of the value.

This page is about an exploratory matrix view. A heatmap does not produce a p-value, effect size, or formal clustering result by itself. Its role is to help you inspect broad patterns across many numeric variables before deciding whether a more specific analysis or figure is needed.

When to use a heatmap

Heatmaps work well when:

• You have five or more numeric columns to compare simultaneously
• You want to see patterns across many variables at once — for example, expression levels across multiple genes or measurements across multiple assay conditions
• The absolute values matter less than the relative pattern across rows and columns

For comparing two variables, a scatter plot is more appropriate. For comparing groups on one variable, a box or violin plot is more appropriate.

The "five or more" guideline is a practical display rule. Below that, readers can usually compare the same variables more directly with simpler figures. The heatmap becomes more useful when the matrix view is doing work that a handful of separate plots would make harder to see.

Required data

A heatmap requires at least five numeric columns. Licklider selects the columns automatically when the data contains enough numeric variables.

That automatic selection is a convenience, not a claim that the chosen variables are the most scientifically important ones. Licklider can detect numeric columns, but it does not determine from the figure alone which measurements are most relevant, which should be excluded for scientific reasons, or which subset best answers the research question.

An optional row label column provides identifiers for each row displayed on the y-axis. Without it, rows are numbered.

Color scale

The heatmap uses the Viridis color scale, which is perceptually uniform and accessible for readers with color vision deficiencies. The scale runs from dark purple (low values) to yellow (high values).

The color scale is fixed and cannot currently be changed to a different palette.

That fixed palette is intentional. Keeping one perceptually uniform scale reduces avoidable visual distortion and makes heatmaps more consistent across datasets and readers. It does not mean that color alone is sufficient to judge biological importance or statistical significance.

Row ordering

Rows can be sorted by their average value across all selected columns. This places rows with consistently high values at one end and rows with consistently low values at the other, making overall patterns easier to see.

Note: The sort-based ordering is not hierarchical clustering. Licklider does not compute linkage distances or dendrogram structures. Rows that appear adjacent are similar in their average magnitude, not necessarily in their pattern across individual columns.

This distinction matters. Two rows can sit next to each other because they have similar overall averages while still differing strongly in which columns are high or low. The current heatmap does not automatically determine whether adjacent rows form a real cluster, share one biological subtype, or have similar multivariable profiles.

If true hierarchical clustering is needed, consider pre-processing the data externally and importing the result with rows already in the desired order.

Normalization

When values across columns are on different scales — for example, one column ranges 0-1 and another ranges 0-1000 — normalization helps make the color encoding comparable across columns.

Normalization is useful when the main question is pattern comparison across columns rather than raw magnitude. It can make relative highs and lows more readable, but it also changes what the colors mean. After normalization, color differences no longer represent the original absolute scale in the same way.

Licklider does not determine automatically which normalization choice is scientifically most appropriate. If absolute values, fold-like changes, or assay-scale units are part of the interpretation, a normalized heatmap can be informative but also easier to over-read if the transformed scale is not kept in mind.

To request normalization, specify it in the Chat:

• "Normalize the heatmap columns"
• "Show z-score normalized values"

Use the heatmap as an exploratory display of matrix structure, not as proof that one cluster, ordering, or normalization view is the uniquely correct one.

What this page does not cover

• PCA for dimensionality reduction → see Choose the Right Figure
• Hierarchical clustering heatmaps (dendrograms, linkage-based ordering) — not currently available in Licklider