claw-metagenomics

What it does

1. Takes paired-end FASTQ files (R1, R2) or a single concatenated FASTQ as input
2. Runs Kraken2 taxonomic classification against a standard database (e.g., Standard-8, PlusPF)
3. Refines abundances with Bracken at species level (read re-estimation)
4. Detects antimicrobial resistance genes with RGI against the CARD database
5. Classifies detected ARGs by WHO critical priority pathogen association
6. Optionally runs HUMAnN3 for functional pathway profiling (MetaCyc + UniRef)
7. Generates three publication-quality figures:
   • Figure 1: Taxonomy bar chart — top 20 species by relative abundance
   • Figure 2: Resistome heatmap — ARG families by drug class with abundance
   • Figure 3: WHO-critical ARG summary — priority-tier breakdown of detected resistance genes
8. Produces a full reproducibility bundle (commands.sh, environment.yml, checksums.sha256)

Why this exists

If you ask a general AI to "analyse a metagenome," it will:

• Not know which Kraken2 database to use or how to set confidence thresholds
• Hallucinate Bracken parameters for read-length and taxonomic level
• Miss the connection between detected ARGs and WHO priority pathogen lists
• Skip HUMAnN3 entirely (or misconfigure its database paths)
• Produce a single bar chart with no resistance context
• Not provide a reproducibility bundle

This skill encodes the correct methodological decisions:

• Kraken2 confidence threshold of 0.2 (reduces false positives in environmental samples)
• Bracken re-estimation at species level with minimum 10 reads
• RGI MAIN with "Perfect" and "Strict" hit criteria only (no "Loose" hits)
• WHO Critical Priority Pathogen list mapped to detected ARG families
• HUMAnN3 with MetaCyc stratification for pathway-level functional context
• Thread count auto-detected from available CPUs
• Full reproducibility bundle for every run

Validated On

The skill works with any shotgun metagenome but has been validated on:

• Peru sewage metagenomics study (6 samples, 3 collection sites: Lima, Cusco, Iquitos)
• Environmental sewage samples with mixed microbial communities
• Read depths ranging from 2M to 15M paired-end reads per sample