Functional Profiling (bioSkills)

A Claude Code skill that quantifies the functional potential of shotgun metagenomes with HUMAnN 3, producing gene-family and pathway abundances stratified by the species that contribute them.


Type	Claude Skill
Supplier	GPTomics bioSkills (community OSS, MIT)
Availability	GA — part of the bioSkills collection
Pricing	Free / OSS (MIT) — HUMAnN, MetaPhlAn, Bowtie2, DIAMOND and the UniRef/ChocoPhlAn databases are separately installed OSS
Capabilities	Read/Write — Claude runs the skill’s workflow locally (Bash/Python), not as an MCP tool
Verified	works · 2026-08-03
Security	cleared · 2026-08-03 — GPTomics/bioSkills MIT confirmed, provenance matches, no advisories

How to install

bioSkills is not an npm package — skills are plain markdown/code read directly by the agent. Clone the repo, then either run the installer for the whole category or copy the single skill directory.

Claude Code — clone and install via the bundled script:
```
git clone https://github.com/GPTomics/bioSkills
cd bioSkills
./install-claude.sh --categories "metagenomics"
```
The installer copies matching skills into ~/.claude/skills/ (default target). Use ./install-claude.sh --list to preview the skills first.
Claude Code / other agents — copy just this one skill:
```
cp -r bioSkills/metagenomics/functional-profiling ~/.claude/skills/
```
(run from inside your clone — the previous step left you in bioSkills/; otherwise replace bioSkills/ with the absolute path of your clone). Install HUMAnN and download the ChocoPhlAn/UniRef databases when prompted on first use.

What it does

Runs HUMAnN 3’s tiered search and interprets the output tables correctly:

Workflow — host depletion and quality trimming (KneadData), a MetaPhlAn 4.1+ taxonomic prescreen that builds a sample-specific ChocoPhlAn pangenome, tier-1 Bowtie2 nucleotide alignment against that pangenome (high-confidence stratification), tier-2 DIAMOND six-frame translated search against UniRef90/50 (inferred stratification), RPK→CPM normalization before any cross-sample statistics, and regrouping to KO/EC/GO with stratified and unstratified tables split out.
Interpretation guidance — a metagenome measures functional potential, not activity (RNA-level validation is required for activity claims); dropping the UNMAPPED/UNINTEGRATED rows biases every downstream comparison; species stratification is an estimate rather than a measurement; coverage vs abundance and the MinPath/gap-fill behaviour; UniRef90-vs-50 and biome-specific database bias.
Alternatives — assembly-based functional annotation, eggNOG-mapper, dbCAN (CAZymes), and antiSMASH (biosynthetic gene clusters).

Primary use cases: MetaCyc pathway abundance across cohorts, gene-family (KO/EC/GO) profiles, choosing read-based vs assembly-based functional annotation.

Notes

Distributed as a SKILL.md (plus reference material) in the bioSkills collection — Claude executes the workflow locally rather than as an MCP server. The upstream skill front-matter name is bio-metagenomics-functional-profiling; if invoked as a namespaced plugin command it resolves under the bioSkills plugin, not as a bare /functional-profiling. HUMAnN’s prescreen step means the catalogued MetaPhlAn Profiling skill covers its taxonomic half. The skill routes AMR-gene questions to AMR Detection and host-gene enrichment to pathway-analysis rather than answering them itself. Upstream directory: metagenomics/functional-profiling.

Sources

Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.