| name | study-design |
| description | Study design for shotgun metagenomics across QC, assembly, binning, and annotation. Use this skill when: (1) Deciding co-assembly vs sample-specific assembly, (2) Defining biological strata (time, treatment, habitat, host), (3) Building a sample/assembly/coverage manifest, (4) Planning multi-assembly binning (e.g. BASALT), (5) Avoiding invalid pooling of unrelated samples. For tool routing, use tool-selection. For stage order, use metagenomics-workflow.
|
| license | MIT |
| category | orchestration |
| tags | ["study-design","co-assembly","pooling","provenance"] |
Study Design
Design principle
Analyse assemblies jointly only when coverage remains biologically interpretable.
Unrelated samples create ambiguous mappings and invalid selection.
Informative multi-sample designs need:
- enough shared genomes for complementary observation
- enough abundance contrast to discriminate contig profiles
Decision table
| Question | If yes | If no |
|---|
| One coherent biological stratum? | May share one recovery run | Split by stratum |
| Coverage traceable to reads + assembly? | Record mapping | Repair provenance first |
| Pooled assembly justified? | Pilot with individuals + matched pool | Prefer sample-specific |
| Pooling erases primary contrast? | Keep strata separate | Pooling may be tested |
When multi-assembly helps
Reactor replicates, longitudinal hosts, nearby sites, matched extraction methods,
before/after intervention, related growth stages, continuous environmental gradients.
Manifest (required fields)
| Field | Purpose |
|---|
sample_id | Stable ID matching filenames |
stratum / Group | Biological partition |
host_reference | Host genome for depletion |
assembly | Contig FASTA |
assembly_type | individual / pool / subgroup |
assembly_reads / coverage_reads | Exact provenance |
run_directory | Dedicated working directory |
Patterns
# Related replicates (feeds stage 3 binning)
sample-specific assemblies ─┐
matched co-assembly ────────┴─> one ensemble binning run (basalt★ / metawrap)
# Longitudinal / treatment
stage1 → run1
stage2 → run2
then compare bins/genomes with declared ANI threshold
Subgroup when samples > ~30 or data > ~300 Gb.
Multi-coverage vs single-coverage binning
Mattock & Watson (Nat. Methods 2023) compared single- vs multi-coverage
binning on the same per-sample assemblies: multi-coverage recovered more
filtered bins and exposed contaminant/chimeric structure that single-coverage
misses (co-abundant contigs within one sample look “genome-like” without
cross-sample contrast).
Practical rule: when related samples exist, map all relevant samples to each
assembly for binning coverage — even if you keep sample-specific assemblies.
Single-coverage is a resource compromise, not a scientific default.
Low-abundance / novelty coassembly
Blind pooling is costly and can worsen assemblies. For large cohorts aimed at
rare/novel genomes, consider binchicken (targeted coassembly via shared
SingleM markers, optional novelty prioritization). Still enforce stratum rules
above — targeting is not a license to mix unrelated biology into one contrast.
Pilot evaluation
Compare designs on retained bins under fixed completeness/contamination,
redundancy (ANI), taxonomy, contiguity, and compute cost — not raw bin count.
Related skills
tool-selection — stage routing; ensemble binning choicemetagenomics-workflow — QC → assembly → binning → taxonomy → function → mining → research-analysisbasalt · metawrap · dastool — binning stagebinchicken — targeted coassembly planning for low-abundance recoverylorbin · comebin · semibin2 · vamb — single binners
