| name | vamb |
| description | Deep variational autoencoder binning of metagenome contigs. Use this skill when working with vamb: installing, choosing parameters, running commands, interpreting outputs, or troubleshooting. Upstream: https://github.com/RasmussenLab/vamb. For stage routing use tool-selection / metagenomics-workflow.
|
| license | MIT |
| category | analysis-tools |
| tags | ["binning","vae","deep-learning"] |
| upstream | https://github.com/RasmussenLab/vamb |
| stage | binning |
VAMB
Upstream: VAMB
Citation
Nissen, J. N. et al. Improved metagenome binning and assembly using deep variational autoencoders. Nat. Biotechnol. 39, 555–560 (2021). https://doi.org/10.1038/s41587-020-00777-4
See also docs/references.md.
Analytical thinking
VAMB encodes coabundance + k-mer features with a VAE, then clusters the
latent space. Key design claims from the paper:
- Integrates composition and coverage without hand-tuned fusion weights.
- Multisplit: cluster across samples, then split into per-sample bins —
aims to use coabundance while avoiding between-sample chimeric genomes that
plague naive coassembly+binning.
- Reported large gains in near-complete genomes on CAMI/MetaHIT-style
benchmarks; strain separation claimed up to ~99.5% ANI in that work.
- Benchmark difficulty (genome entropy) correlated with VAMB’s relative gains —
expect more benefit in complex communities, not a flat uplift everywhere.
Use VAMB as a diverse ensemble member vs MetaBAT2/MaxBin2. Still run
CheckM2; prefer basalt ★ / metawrap / dastool for multi-binner consensus
when publishing catalogues.
How to run
conda install -c bioconda vamb
vamb --outdir temp/bins/vamb --fasta contigs.fa --bamfiles temp/map/*.bam -p 16
Key parameters
| Flag | Notes |
|---|
--bamfiles | Sorted BAMs; multi-sample improves coabundance signal |
| multisplit / clustering | Follow upstream defaults unless reproducing a paper |
| GPU | Optional acceleration |
Literature notes
- Do not confuse VAMB’s “multisplit bins” with BASALT multi-assembly refinement.
- Adding VAMB bins into DAS Tool ensembles was beneficial in the paper’s
comparisons — diversity of algorithms matters.
- Closely related strains may still collapse; validate with
skani / instrain.
Related skills
semibin2 · comebin · metabat2 · dastool · basalt · skani ·
tool-selection · metagenomics-workflow
