Skill: metagenome-host-removal

Use When

• The user has shotgun metagenomic reads contaminated with host DNA.
• The user wants to remove human, mouse, or other host reads before taxonomic or assembly analysis.
• The user needs to quantify the host contamination rate in a metagenomic sample.

Inputs

• Required:
  • FASTQ file(s) — single-end or paired-end reads (.fastq, .fq, .fastq.gz, .fq.gz).
  • Host reference genome index — pre-built Bowtie2 or BWA-MEM2 index.
• Optional:
  • --aligner — Aligner to use (bowtie2 or bwa-mem2, default: bowtie2).
  • --threads N — Number of threads (default: 4).
  • --outdir DIR — Output directory (default: host_removal_results).
  • --sensitivity — Bowtie2 sensitivity preset (--very-sensitive, --sensitive; default: --very-sensitive).

Workflow

1. Align reads to the host reference genome using Bowtie2 or BWA-MEM2.
2. Extract unmapped reads (samtools view -f 4 for SE; -f 12 for PE unmapped pairs).
3. Sort unmapped reads by name (samtools sort -n).
4. Convert back to FASTQ (samtools fastq).
5. Report: total reads, host reads removed, non-host reads retained, host contamination rate.

Output Contract

• Host-depleted FASTQ(s) — One or two FASTQ files with host reads removed (<outdir>/<sample>_hostdepleted_R1.fastq.gz, _R2.fastq.gz for PE).
• Host contamination summary — Total reads, host reads, non-host reads, host contamination percentage.

Limits

• Bowtie2 or BWA-MEM2 must be installed and available on $PATH.
• samtools must be installed and available on $PATH.
• Host genome index must be pre-built (use bowtie2-build or bwa-mem2 index).
• Human reference: GRCh38 (hg38) is recommended for human host removal.
• For clinical samples, host removal rate should be >95%; lower rates suggest index mismatch or sample issues.
• This skill does not build the host index; the user must provide a pre-built index path.
• Common failure cases:
  • Host genome index path incorrect or index files incomplete, causing aligner to abort.
  • Using the wrong host reference genome (e.g., mouse index for a human-derived sample).
  • Paired-end read extraction flags incorrect, producing orphan reads without mates.