| name | mag-qc |
| description | Quality thresholds after binning (completeness/contamination, dRep ANI, publication vs refinement gates). Use this skill when: (1) Setting filters for bins before calling them MAGs, (2) Choosing dRep ANI cutoffs, (3) Interpreting CheckM2 reports, (4) Separating refinement entry gates from publication filters, (5) Ranking bin catalogues, (6) Choosing Deepurify vs Anvi'o vs GUNC. For binning, use basalt (recommended), metawrap, or dastool.
|
| license | MIT |
| category | evaluation |
| tags | ["qc","checkm2","drep","completeness","contamination"] |
MAG Quality Control
Critical limitation
Marker/ML completeness/contamination are estimates, not ground truth —
especially for unusual lineages. Treat them as filters and reporting metrics.
Quick thresholds
| Criterion | Common working gate | Stricter publication |
|---|
| Completeness | ≥ 50% (MetaWRAP refine) / ≥ 35% (BASALT entry) | ≥ 90% (near-complete) |
| Contamination | ≤ 10% refine / ≤ 20% BASALT entry | ≤ 5% |
| dRep species ANI | 95% | study-defined |
| dRep strain ANI | 99% | study-defined |
Declare thresholds before inspecting results. Do not optimize for maximum bin count.
Keep refinement entry gates distinct from publication filters in Methods.
Recommended QC narrative
bins
→ checkm2 (comp/cont scores)
→ gunc (chimerism / clade contamination)
→ optional deepurify (remove/split) or anvio (interactive) → re-checkm2/gunc
→ drep (skani|fastani ANI) → species/strain reps
→ gtdbtk / mining
Tool roles
| Tool | Role |
|---|
checkm2 | ML completeness / contamination scores |
gunc | Chimeric / foreign-lineage contamination |
deepurify | Automated contig removal / sub-MAG split |
anvio | Human-guided contig refinement |
drep | Dereplicate by ANI |
skani / fastani | ANI engines / cross-checks |
busco | Optional gene-set view (not a CheckM2 replacement) |
CheckM2
checkm2 predict --input bins/ --output-directory checkm2_out --threads 16
Record database path/version. Do not silently mix CheckM1 and CheckM2 across
design comparisons.
Scoring / ranking (example)
- Pass hard filters (comp/cont/size)
- Prefer higher completeness, lower contamination
- Prefer better contiguity and read support
- Prefer taxonomic consistency; discordant GUNC → review
Related skills
basalt · metawrap · dastool · checkm2 · gunc · deepurify ·
anvio · drep · skani · fastani · busco · visualization ·
metagenomics-workflow
