Skip to content

Using DeepMapper with Claude

The repository ships a Claude Code skill at .claude/skills/deepmapper/SKILL.md. When you open this repo in Claude Code, the skill is available and Claude can drive a full DeepMapper analysis for you: load the data, run a fast linear check, run the pipeline, read the gene chord, and reproduce an example figure.

How it works

A Claude skill is a Markdown file with a short description in its front matter. Claude reads the description to decide when the skill applies, then follows the workflow in the body. The DeepMapper skill triggers when you ask Claude to:

  • analyse a single-cell or omics matrix without dimension reduction,
  • find the genes (or gene chords) that separate cell states,
  • attribute a classifier back to named features,
  • or reproduce one of the example analysis figures.

Use it

  1. Install Claude Code and open this repository in it.
  2. Ask in plain language, for example:
  3. "Analyse this h5ad with DeepMapper and tell me which genes separate the states."
  4. "Find the gene chord for CD4 memory in this matrix."
  5. "Reproduce the Figure 1 analysis from the bench scripts."
  6. Claude loads the skill and walks the workflow: it checks the install, loads your data with pydeepmapper.io, runs a deterministic linear baseline first, then runs pydeepmapper.runner.run, and reads findings.ranking(...).

You can also invoke it explicitly by name with /deepmapper if your Claude Code setup lists it.

What the skill will and will not do

The skill follows the same honest-analysis rules:

  • It keeps every feature. It will not select highly variable genes or run PCA first.
  • It runs the fast, deterministic linear baseline before the CNN, and expects the two rankings to roughly agree.
  • It reports a finding as a module (a set of interchangeable genes), not a single fixed gene list, and it will not over-read a single run.
  • It uses cnn_small by default and avoids resnet18 on sparse single-cell data.

Extending the skill

The skill is plain Markdown. Edit .claude/skills/deepmapper/SKILL.md to add your own datasets, default configs, or house rules. Keep the front-matter description precise, since that is what Claude matches against your request.