How to bin MAGs

Step 1 — Index contigs with `index-contigs`¶

Build a Bowtie2 index for each sample’s contigs using the assembly plugin. This index enables read mapping in the next step and is required by both binners.

With parsl parallelization

Without parallelization

mosh assembly index-contigs \
    --i-contigs cache:contigs \
    --p-threads 8 \
    --p-seed 100 \
    --o-index cache:contigs_index \
    --parallel-config parallel.config.toml \
    --verbose

Step 2 — Map reads to contigs with `map-reads`¶

Map the original reads back to the assembled contigs. Both binners use the resulting alignment maps to estimate contig coverage across samples, which is a key signal for grouping contigs into bins.

With parsl parallelization

Without parallelization

mosh assembly map-reads \
    --i-index cache:contigs_index \
    --i-reads cache:reads \
    --p-threads 8 \
    --p-seed 100 \
    --o-alignment-maps cache:reads_to_contigs_aln \
    --parallel-config parallel.config.toml \
    --verbose

Step 3 — Bin contigs into MAGs¶

Two binners are available in the mag plugin. Both take the same contigs and alignment maps as input; choose one based on your environment and expected sample type.

MetaBAT 2

SemiBin2

bin-contigs-metabat uses tetranucleotide frequency together with coverage information to group contigs. It is fast, well-established, and explicitly outputs the contigs that could not be assigned to any bin (unbinned-contigs), which is used later in the BUSCO visualization.

mosh mag bin-contigs-metabat \
    --i-contigs cache:contigs \
    --i-alignment-maps cache:reads_to_contigs_aln \
    --p-num-threads 4 \
    --p-seed 100 \
    --o-mags cache:mags \
    --o-contig-map cache:contig_map \
    --o-unbinned-contigs cache:unbinned_contigs \
    --verbose

bin-contigs-semibin2 uses deep learning together with coverage information. Pre-trained environment models are available for common sample types (human_gut, ocean, soil, wastewater, and more). For samples from an environment not covered by a pre-trained model, use global. Unlike MetaBAT 2, SemiBin2 does not produce an unbinned-contigs artifact.

mosh mag bin-contigs-semibin2 \
    --i-contigs cache:contigs \
    --i-alignment-maps cache:reads_to_contigs_aln \
    --p-environment global \
    --p-training-type semi \
    --p-threads 4 \
    --p-random-seed 100 \
    --o-mags cache:mags \
    --o-contig-map cache:contig_map \
    --verbose

Replace global with the environment that best matches your samples (e.g., human_gut, soil, ocean) for improved binning accuracy.

MetaBAT 2 vs SemiBin2 — when to use which¶

Aspect	MetaBAT 2	SemiBin2
Algorithm	Tetranucleotide frequency + coverage	Deep learning + coverage
Speed	Fast	Slower (especially without GPU)
Environment-specific models	No	Yes — improves accuracy for known environments
Unbinned contigs output	Yes	No
Best for	Any dataset; good default	Samples from well-characterized environments

Step 4 — Evaluate MAG quality with `evaluate-busco`¶

BUSCO assesses completeness and contamination of each MAG by checking for the presence of lineage-specific single-copy marker genes. First download the relevant BUSCO database with fetch-busco-db:

mosh mag fetch-busco-db \
    --p-lineages bacteria_odb12 \
    --o-db cache:busco_db \
    --verbose

Common lineages: bacteria_odb12, archaea_odb12, eukaryota_odb12. Use bacteria_odb12 for typical gut/soil/environmental metagenomes. Then run the evaluation:

With parsl parallelization

Without parallelization

mosh mag evaluate-busco \
    --i-mags cache:mags \
    --i-db cache:busco_db \
    --i-unbinned-contigs cache:unbinned_contigs \
    --p-lineage-dataset bacteria_odb12 \
    --p-cpu 4 \
    --o-results cache:busco_results \
    --o-visualization mags.qzv \
    --parallel-config parallel.config.toml \
    --verbose

mosh mag evaluate-busco \
    --i-mags cache:mags \
    --i-db cache:busco_db \
    --i-unbinned-contigs cache:unbinned_contigs \
    --p-lineage-dataset bacteria_odb12 \
    --p-cpu 4 \
    --o-results cache:busco_results \
    --o-visualization mags.qzv \
    --verbose

Alternative: CheckM

q2-checkm provides genome quality assessment using CheckM v1, an earlier but widely used marker-gene approach. It requires a separate installation:

qiime checkm evaluate-bins \
    --i-bins cache:mags \
    --p-threads 4 \
    --o-visualization checkm-results.qzv \
    --verbose

CheckM and BUSCO use different marker gene sets. BUSCO’s ODB12 databases are more comprehensive and better maintained for environmental metagenomes.

Step 5 — Filter MAGs by quality with `filter-mags`¶

Remove low-quality bins before dereplication or downstream analyses. The MIMAG standard defines “medium quality” as ≥50% completeness and <10% contamination:

mosh mag filter-mags \
    --i-mags cache:mags \
    --m-metadata-file cache:busco_results \
    --p-where "completeness>50 AND contamination<10" \
    --p-on "mag" \
    --o-filtered-mags cache:mags_filtered \
    --verbose

Adjust thresholds based on your downstream goals. For high-confidence phylogenomic analyses you may require ≥90% completeness and <5% contamination (“high quality” by MIMAG). For broad community profiling, more permissive thresholds may be appropriate.

After dereplication, you can apply a second round of filtering on the dereplicated set with filter-derep-mags, which accepts a FeatureData[MAG] artifact instead of SampleData[MAGs].

QC checkpoint¶

In the BUSCO visualization (mags.qzv), inspect:

Completeness distribution — most MAGs from a well-assembled community should be >50% complete.
Contamination — bins with >10% contamination likely contain sequences from multiple organisms; consider discarding them.
Unbinned contig fraction (MetaBAT 2 only) — a large unbinned fraction suggests many contigs were too short or had insufficient coverage to be confidently assigned; this is normal for diverse or low-coverage samples.

Step 1 — Index contigs with index-contigs¶

Step 2 — Map reads to contigs with map-reads¶

Step 3 — Bin contigs into MAGs¶

MetaBAT 2 vs SemiBin2 — when to use which¶

Step 4 — Evaluate MAG quality with evaluate-busco¶

Step 5 — Filter MAGs by quality with filter-mags¶

QC checkpoint¶

Further reading¶

Step 1 — Index contigs with `index-contigs`¶

Step 2 — Map reads to contigs with `map-reads`¶

Step 4 — Evaluate MAG quality with `evaluate-busco`¶

Step 5 — Filter MAGs by quality with `filter-mags`¶