Free AI Heatmap Generator for scientific data

Upload the expression matrix and ask for, for example, “row z-scores, cluster genes and samples, show dendrograms and a condition bar”. FigCanvas writes and runs the Python code on your data, so the clustering is computed rather than drawn by hand, and you can ask for changes in the same chat. It is the figure pheatmap(mat, scale = "row") draws in R or seaborn.clustermap in Python, without installing either.

Yes, as long as it computes from your numbers rather than drawing an image. An image generator invents the cells; FigCanvas reads your file, computes z-scores, distances and clustering in Python and colors each cell from those values, so the heatmap matches the data and can be reproduced.

A wide matrix: the first column holds row labels such as gene symbols, and each other column holds one sample, for example normalized counts or TPM exported from DESeq2 or edgeR. For a correlation heatmap, upload one row per subject and one numeric column per variable. CSV and Excel files both work, and a long table with gene, sample and value columns can be pivoted for you.

Yes; ask for it in the prompt. For expression data, apply log2(x + 1) to counts or TPM, then scale each row to a z-score, (value − row mean) / row SD, so every gene uses the same color range, and clip the scale at about ±2. Hierarchical clustering is standard: pheatmap defaults to Euclidean distance with complete linkage, while correlation distance (1 − Pearson r), used in the example above with average linkage, groups genes by the shape of their profile. Drop near-constant rows first; 6 of the 32 genes in the example were housekeeping genes with almost no variance.

Yes. Upload one column per variable and ask for Pearson r (roughly linear data) or Spearman ρ (ranks and skewed data), shown in one triangle on a diverging scale from −1 to 1 with r printed in each cell. Ask for a multiple-testing correction too: 11 variables give 55 pairs, and with Bonferroni and 240 subjects only |r| ≥ 0.21 reaches P < 0.05, as in the clinical example above.

Yes; name it in the prompt. Viridis suits sequential data and blue-white-red (ColorBrewer RdBu) suits diverging data such as z-scores; both read well with the common forms of color vision deficiency. Avoid red-green and rainbow (jet) scales: red-green color vision deficiency affects about 8% of men and 0.5% of women of Northern European descent, and rainbow scales add false boundaries.

Pro and Credit Pack users export editable SVG and 300 DPI PDF, so labels and fonts can be adjusted in Illustrator or Inkscape to the journal’s guidelines; free exports are watermarked PNG. If the journal asks for TIFF, export a PNG and convert it with the free FigCanvas PNG to TIFF converter.

It depends on where your data already lives. In R, pheatmap and ComplexHeatmap are the standard for clustered and annotated heatmaps; in Python, seaborn.clustermap. Free web tools draw heatmaps without code: Morpheus from the Broad Institute and SRplot cluster an uploaded matrix, and Heatmapper2 displays expression data, each with fixed layout options. FigCanvas takes a CSV or Excel file and a description in plain language, and returns a clustered or correlation heatmap you can export as editable SVG or PDF.

Only a simple one. Select the numbers and use Home > Conditional Formatting > Color Scales; Excel colors each cell by its value. It cannot cluster rows or columns, draw dendrograms or scale rows to z-scores, so those steps need extra formulas and manual sorting. To skip that, upload the same sheet to the FigCanvas heatmap generator.