deepurify

name	deepurify
description	Deepurify — multi-modal deep language model for MAG decontamination by matching contigs to taxonomic lineages and splitting chimeric bins into higher-quality sub-MAGs. Use when: (1) Cleaning contaminated MAGs after binning, (2) Rescuing near-threshold contaminated bins, (3) Complementing CheckM2/GUNC (detection vs contig removal). Upstream: https://github.com/ericcombiolab/Deepurify.
license	MIT
category	analysis-tools
tags	["decontamination","MAG","Deepurify","QC"]
upstream	https://github.com/ericcombiolab/Deepurify
stage	assessment

Deepurify

Upstream: ericcombiolab/Deepurify

Citation

Zou, B. et al. A multi-modal deep language model for contaminant removal from metagenome-assembled genomes. Nat. Mach. Intell. 6, 1245–1255 (2024). https://doi.org/10.1038/s42256-024-00908-5

See also docs/references.md.

Analytical thinking

CheckM2/GUNC score contamination; Deepurify aims to remove contaminant contigs and split chimeras into sub-MAGs using sequence↔taxonomy multimodal embeddings. Paper reports substantial gains in high-quality MAG counts across soil, ocean, plant, freshwater, and human faecal sets — treat as benchmark-conditioned.

Recommended order:

bins → checkm2 (± gunc)
    → deepurify on contaminated / borderline bins
    → re-QC with checkm2 / gunc
    → drep / gtdbtk

Do not skip re-QC after decontamination. Model errors can drop true contigs — inspect high-value novel lineages manually (anvio).

How to run

# verify CLI in upstream README for your version
deepurify \
  -i bins/ \
  -o temp/deepurify \
  # model / DB paths per docs

Decision tree

MAG contamination problem?
├─ Detect / filter only → checkm2 / gunc / mag-qc
├─ Remove contigs / split chimeras → deepurify
├─ Interactive contig surgery → anvio
└─ Prevent upstream → multi-coverage binning + ensemble (basalt ★)

Related skills

checkm2 · gunc · mag-qc · anvio · basalt · drep · tool-selection