Variational Autoencoders for Metagenomic Binning (Vamb)#
Vamb is a family of metagenomic binners which feeds kmer composition and abundance into a variational autoencoder and clusters the embedding to form bins. Its binners perform excellently with multiple samples, and pretty good on single-sample data.
Programs in Vamb#
The Vamb package contains several programs, including three binners:
TaxVamb: A semi-supervised binner that uses taxonomy information from e.g.
mmseqs taxonomy. TaxVamb produces the best results, but requires you have run a taxonomic annotation workflow. Link to article.Vamb: The original binner based on variational autoencoders. This has been upgraded significantly since its original release. Vamb strikes a good balance between speed and accuracy. Link to article.
Avamb: An obsolete ensemble model based on Vamb and adversarial autoencoders. Avamb has an accuracy in between Vamb and TaxVamb, but is more computationally demanding than either. We don’t recommend running Avamb: If you have the compute to run it, you should instead run TaxVamb See the Avamb README page for more information. Link to article.
And a taxonomy predictor:
Taxometer: This tool refines arbitrary taxonomy predictions (e.g. from
mmseqs taxonomy) using kmer composition and co-abundance. Link to article
See also our tool BinBencher.jl for evaluating metagenomic bins when a ground truth is available, e.g. for simulated data or a mock microbiome.