You have run your binning pipeline and recovered a set of metagenome-assembled genomes (MAGs). Now comes a critical question: how good are they, and which ones can you actually use for downstream analysis?
MAG quality assessment is not just a formality before moving on. It determines which bins are worth annotating, which can be included in comparative analyses, and which should be discarded or flagged as unreliable. This guide explains how to assess MAG quality using CheckM2 and GUNC, what each metric means, and how to apply the MIMAG standards to classify your results. For a broader look at why MAGs become low quality in the first place, see our post on Low-Quality MAGs: Common Causes and Fixes.
If you need support with MAG quality assessment or full metagenomics analysis, explore our Metagenomics Bioinformatics Services — we work with data from any sequencing platform, no wet-lab work involved.
Why MAG quality assessment matters
MAGs are reconstructed computationally from mixed community sequencing data. Unlike isolate genomes from pure cultures, they are not guaranteed to be complete, uncontaminated, or derived from a single organism. Errors introduced during assembly, binning, or bin refinement can produce bins that appear plausible but contain sequences from multiple organisms, miss key genomic regions, or have inflated genome size estimates.
Downstream analyses — metabolic reconstruction, comparative genomics, phylogenomics, functional annotation — all depend on the quality of the MAGs you use as input. A contaminated MAG can introduce false gene presence. An incomplete MAG can produce a misleading picture of metabolic potential. Quality assessment is the checkpoint that separates usable bins from noise.
The two core metrics: completeness and contamination
Both CheckM and CheckM2 estimate MAG quality based on single-copy marker genes — conserved genes expected to be present exactly once in nearly all bacterial or archaeal genomes. The logic is straightforward:
- Completeness (%) — the proportion of expected single-copy marker genes that are present in your MAG. A completeness of 90% means that 90% of the expected markers were found. Missing markers suggest the genome is incomplete.
- Contamination (%) — the proportion of marker genes present in more than one copy. Duplicated single-copy genes suggest that sequences from more than one organism ended up in the same bin.
These two metrics are complementary. A bin can have high completeness but also high contamination (suggesting a chimeric genome), or low completeness with low contamination (suggesting an incomplete but clean bin). Both numbers are needed to interpret quality.
CheckM vs CheckM2: which one to use?
The original CheckM uses lineage-specific marker gene sets and has been the standard tool for MAG quality assessment since 2015. CheckM2 was published in 2022 and uses machine learning to predict completeness and contamination, making it substantially faster, more accurate for novel lineages, and better suited to diverse or poorly characterised organisms such as those found in the Patescibacteria or DPANN superphylum.
| Feature | CheckM (v1) | CheckM2 |
|---|---|---|
| Method | Lineage-specific marker gene sets | Machine learning (gradient boosting) |
| Speed | Slower | Substantially faster |
| Novel lineages | Less accurate for sparse lineages | Better performance across the tree of life |
| Reduced-genome organisms (e.g. symbionts) | Can underestimate completeness | More reliable estimates |
| Database updates | Static marker sets | Updatable with new reference genomes |
| Current recommendation | Still widely used; required by some pipelines | Recommended for new projects |
For new projects in 2026, CheckM2 is the recommended choice. Many published protocols now report CheckM2 results as the primary quality metric, sometimes alongside CheckM1 for comparison. If you are submitting genomes to NCBI or following a protocol that specifies CheckM1, use whichever is required — but understand that the two tools can produce different completeness and contamination estimates for the same bin.
Running CheckM2 and reading the output
A basic CheckM2 run on a directory of MAG FASTA files looks like this:
checkm2 predict --threads 16 --input /path/to/mag/fasta/ --output-directory /path/to/checkm2_output/ -x faThe main output is a tab-separated file quality_report.tsv with one row per MAG. The key columns to focus on:
| Column | What it means | What to look for |
|---|---|---|
| Completeness | % of expected marker genes present | >90% = high quality; 50–90% = medium quality; <50% = low quality |
| Contamination | % of marker genes duplicated | <5% = high quality; <10% = medium quality; >10% = problematic |
| Completeness_Model_Used | Which ML model was applied | General model vs lineage-specific; check for novel lineages |
| Translation_Table_Used | Genetic code applied | Non-standard codes (e.g. 4, 25) indicate unusual organisms; verify |
| Genome_Size | Total assembly size in base pairs | Compare to expected range for the taxon; very large or small values are a flag |
| GC_Content | GC% of the MAG | Cross-check with the expected range for the taxonomic group |
| Contig_N50 | Contig continuity metric | Higher N50 = more contiguous assembly; relevant for downstream tools |
| Coding_Density | Proportion of genome that encodes genes | Unusually low coding density may indicate contamination or assembly artefacts |
MIMAG quality standards: how to classify your MAGs
The Minimum Information about a Metagenome-Assembled Genome (MIMAG) framework provides the community standard for reporting MAG quality in publications and database submissions. The thresholds are widely adopted across journals and sequencing repositories.
| MIMAG tier | Completeness | Contamination | Additional criteria | Suitable for |
|---|---|---|---|---|
| High quality (HQ) | >90% | <5% | 5S, 16S, 23S rRNA genes present; ≥18 tRNA genes | Detailed metabolic reconstruction, comparative genomics, NCBI submission, phylogenomics |
| Medium quality (MQ) | ≥50% | <10% | None beyond completeness/contamination | Exploratory analysis, functional inference, taxonomic placement |
| Low quality (LQ) | <50% | Any | — | Very limited; may identify dominant organisms or specific genes; interpret cautiously |
In practice, most published metagenomics studies report only HQ and MQ MAGs and discard or set aside LQ bins. A useful quality score for ranking MAGs within a tier is the Parks score: completeness − (5 × contamination). MAGs with a score below 30–50 are often excluded from downstream analysis even if they nominally meet MQ thresholds.
Checking for rRNA genes and tRNAs (HQ requirement)
CheckM2 does not check for rRNA genes or tRNA count — those are required for MIMAG HQ classification but assessed separately. Use Barrnap for rRNA detection and tRNAscan-SE for tRNA annotation. A MAG that passes CheckM2 thresholds (>90% completeness, <5% contamination) may still fail HQ classification if it lacks rRNA genes — which is common in fragmented assemblies.
GUNC: detecting chimerism that CheckM2 misses
CheckM2 is excellent at estimating completeness and gross contamination, but it has a known limitation: it can miss chimeric MAGs — bins composed of sequences from multiple organisms that share similar marker genes. A chimeric MAG may show low CheckM2 contamination (because duplicated single-copy genes are not present) while still containing contigs from completely different taxa.
GUNC (Genome UNClutterer) addresses this directly. Rather than counting duplicated marker genes, GUNC measures the taxonomic consistency of individual contigs using a large reference gene database. If your MAG’s contigs map to genes from very different lineages, GUNC flags it as chimeric — even if CheckM2 reports low contamination.
A conservative estimate suggests that 15–30% of pre-filtered “high quality” MAGs in published studies contain undetected chimerism that GUNC would flag. In public databases (GenBank, RefSeq), approximately 5–6% of genomes show this type of contamination.
Key GUNC output metrics
The main GUNC output is GUNC.progenomes_2.1.maxCSS_level.tsv. The most important columns are:
| Column | What it means | Threshold |
|---|---|---|
| pass.GUNC | TRUE/FALSE: does the MAG pass GUNC quality check? | TRUE = passes; FALSE = suspected chimera |
| clade_separation_score (CSS) | Degree of taxonomic separation between contigs | CSS < 0.45 generally considered acceptable; higher values suggest chimerism |
| contamination_portion | Estimated fraction of genes from foreign lineages | Lower is better; context-dependent |
| n_effective_surplus_clades | Number of excess taxonomic groups detected | 0 = clean; >0 = multiple lineages present |
| reference_representation_score | How well the organism is represented in the reference database | Low score = poorly characterised lineage; results less reliable |
A basic GUNC run against the ProGenomes database:
gunc run --input_dir /path/to/mag/fasta/ --out_dir /path/to/gunc_output/ --db_file /path/to/gunc_db_progenomes2.1.dmnd --threads 16 -rUsing CheckM2 and GUNC together
The recommended workflow is to run both tools and combine their outputs before deciding which MAGs to take forward. A MAG should pass both CheckM2 thresholds and GUNC to be considered reliable.
| CheckM2 result | GUNC result | Interpretation | Action |
|---|---|---|---|
| HQ (>90% / <5%) | pass.GUNC = TRUE | High-quality, non-chimeric MAG | Proceed to annotation and downstream analysis |
| HQ (>90% / <5%) | pass.GUNC = FALSE | Chimeric despite apparent low contamination | Inspect contigs; attempt bin refinement; flag in reporting |
| MQ (50–90% / <10%) | pass.GUNC = TRUE | Incomplete but clean MAG | Suitable for exploratory analysis; report limitations |
| MQ (50–90% / <10%) | pass.GUNC = FALSE | Incomplete and potentially chimeric | Attempt refinement; treat with caution |
| LQ (<50%) | Any | Incomplete bin | Set aside unless specific genes of interest are confirmed |
Combining both tools is now considered best practice in published metagenomics workflows. Some automated pipelines — including nf-core/mag — run both tools as part of the standard MAG quality control step.
Additional metrics to inspect beyond completeness and contamination
Completeness and contamination are the primary filters, but a complete quality assessment should also consider:
- Number of contigs. A MAG with high completeness but 800 contigs will be harder to annotate reliably and may have lower marker gene confidence than one with 20 contigs. Long-read or hybrid assemblies often produce fewer, more contiguous bins.
- Contig N50. Higher N50 means longer contigs on average, which improves gene prediction, operon detection, and downstream tools that depend on contig length.
- Genome size. Compare to expected genome size for the taxonomic group. An unusually small genome (e.g. <1 Mb for a typically 4 Mb organism) may indicate an incomplete bin. An unusually large genome may suggest contamination that marker-gene methods missed.
- GC content. Should be consistent across contigs within the bin and plausible for the assigned taxon. Highly variable GC content within a single MAG is a red flag.
- Coding density. Most bacterial genomes have 85–92% coding density. Values outside this range deserve inspection.
- Taxonomic consistency (GTDB-Tk). Running GTDB-Tk on your MAGs provides taxonomic placement and an additional internal consistency check. A MAG with mixed taxonomic signals at the contig level is a further chimerism indicator.
MAG dereplication: removing redundancy before downstream analysis
If you have assembled and binned multiple samples (or co-assemblies), you will likely recover the same organism multiple times. Before downstream analysis, redundant MAGs representing the same species should be collapsed into a single representative genome. This step is called dereplication.
The standard tool is dRep, which first clusters MAGs at a coarse similarity threshold (typically 90% ANI for primary clusters) and then applies a finer 95% ANI threshold for species-level dereplication. Within each cluster, the representative MAG is selected based on a quality score: completeness − (5 × contamination) + 0.5 × log(N50).
A standard dRep command filtering to MQ+ MAGs before dereplication:
dRep dereplicate /path/to/drep_output/ \
-g /path/to/mag/fasta/*.fa \
-comp 50 \
-con 10 \
-sa 0.95 \
-nc 0.4 \
--genomeInfo checkm2_quality.csvThe --genomeInfo file should contain CheckM2 completeness and contamination values per MAG, formatted as a CSV with columns: genome, completeness, contamination. This ensures dRep selects the best representative based on quality rather than arbitrary order.
For the full upstream workflow that produces the MAGs you are now assessing, see our Metagenome Assembly Pipeline guide and the Metagenomic Binning Tools comparison.
Interpreting quality metrics for poorly characterised lineages
Both CheckM2 and GUNC rely on reference databases. For well-characterised phyla (Firmicutes, Proteobacteria, Bacteroidetes), predictions are generally reliable. For novel or poorly characterised lineages — particularly in the Patescibacteria, DPANN Archaea, or deep-branching bacterial phyla — quality estimates may be less accurate:
- CheckM2 may underestimate completeness for symbionts and reduced-genome organisms, as many expected marker genes are genuinely absent in these lineages rather than missing from the bin.
- GUNC’s
reference_representation_scoreprovides a built-in flag: a low score means the organism is poorly represented in the ProGenomes database, so the chimerism assessment is less certain. - For novel lineages, cross-referencing with GTDB-Tk placement and manual inspection of contig coverage and GC homogeneity provides additional confidence.
A practical MAG quality assessment checklist
- Run CheckM2 on all bins; record completeness, contamination, genome size, GC content, N50 and coding density
- Apply MIMAG thresholds to classify bins as HQ, MQ or LQ
- For HQ candidates, confirm rRNA genes (Barrnap) and tRNA count (tRNAscan-SE)
- Run GUNC on all MQ+ bins; exclude or flag those that fail
- Calculate the Parks quality score (completeness − 5 × contamination) and exclude bins below your study threshold (commonly 50)
- Inspect GC content and coverage consistency for bins near thresholds
- Run GTDB-Tk for taxonomic placement and additional consistency checks
- Run dRep at 95% ANI to dereplicate redundant MAGs before downstream analysis
- Report completeness, contamination, GUNC pass/fail, N50, contig count and MIMAG tier for all retained MAGs
Frequently asked questions
What completeness threshold should I use for downstream analysis?
This depends on the biological question. For detailed metabolic reconstruction and phylogenomics, use HQ MAGs (>90% completeness, <5% contamination). For exploratory taxonomic profiling or identifying dominant organisms, MQ MAGs (≥50% completeness, <10% contamination) are often sufficient. Always report what threshold you applied and why.
My MAG has high contamination but high completeness — is it useful?
High contamination is generally more problematic than low completeness. A contaminated bin contains sequences from multiple organisms, which can produce false signals in metabolic analysis, gene presence/absence comparisons, and phylogenetic placement. Try bin refinement tools (e.g. anvi’o) before discarding a high-completeness but contaminated MAG.
Can I use CheckM2 for Archaea?
Yes. CheckM2 was trained on both bacterial and archaeal genomes and works across the prokaryotic tree of life. However, accuracy is lower for Archaea from poorly characterised lineages (e.g. DPANN), where reduced genomes mean many expected marker genes are genuinely absent.
Should I always run GUNC?
For any project where MAG quality will be reported or where MAGs are used for metabolic or comparative analysis, yes. GUNC adds a relatively small computational cost but catches chimeric MAGs that CheckM2 misses. It is particularly important when assemblies are highly fragmented or when the community contains many closely related organisms.
What is a good number of MAGs to expect from a metagenomics project?
This varies enormously by sample type, sequencing depth, community complexity, and platform. Simple communities (bioreactors, enrichment cultures) may yield dozens of HQ MAGs with 10 Gb of data. Complex soil or sediment metagenomes may yield only a handful of HQ MAGs even with 50 Gb. Setting realistic expectations before the project starts avoids misinterpreting technically sound results as a pipeline failure.
Need help assessing or interpreting your MAGs?
Tailoredomics provides downstream metagenomics bioinformatics analysis — no wet-lab, no sequencing. Whether you have raw FASTQ files or assembled bins ready for quality assessment, we handle the full workflow: assembly, binning, CheckM2/GUNC quality control, GTDB-Tk classification, dereplication, functional annotation, and interpretation.
➜ Explore our Metagenomics Bioinformatics Services or get in touch to discuss your dataset.