| name | lorbin |
| description | LorBin — unsupervised long-read metagenome binning with multiscale adaptive clustering (DBSCAN + BIRCH) and single-copy-gene evaluation. Use when: (1) Binning long-read / HiFi metagenome assemblies, (2) Recovering MAGs from species-rich or abundance-imbalanced communities, (3) Comparing LorBin to SemiBin2/VAMB long-read modes, (4) Feeding LorBin bins into ensemble refinement (BASALT `-e l` needs the PKU-EMBL LorBin fork). Upstream: https://github.com/LorMeBioAI/LorBin. For ensemble recovery use basalt. For short-read defaults use metabat2 / semibin2 / vamb.
|
| license | MIT |
| category | analysis-tools |
| tags | ["binning","long-read","LorBin","MAG","DBSCAN"] |
| upstream | https://github.com/LorMeBioAI/LorBin |
| stage | binning |
LorBin
Upstream: LorMeBioAI/LorBin
Citation
Xue, W. et al. LorBin: efficient binning of long-read metagenomes by
multiscale adaptive clustering and evaluation. Nat. Commun. 16, 9353
(2025). https://doi.org/10.1038/s41467-025-64916-8
See also docs/references.md.
Analytical thinking
Long-read assemblies differ from short-read graphs: fewer, longer contigs,
sharper abundance separation, and often unknown / imbalanced taxa. Short-read
binners (and some DL tools trained mainly on short-read CAMI) under-recover
novel lineages in that regime.
LorBin is an unsupervised long-read binner: multiscale adaptive DBSCAN then
BIRCH, with iterative single-copy-gene evaluation to accept/recluster bins. The
paper reports large gains in high-quality MAGs and novel taxa vs several
competitors on oral, gut, and marine long-read sets — treat those as
benchmark-conditioned, not universal guarantees.
Within this package:
| Role | Skill |
|---|
| Long-read single binner | lorbin (or SemiBin2 long-read mode) |
| Short-read single binners | metabat2 · semibin2 · vamb · comebin |
| Ensemble refinement | basalt ★ · metawrap · dastool |
If using BASALT with LorBin as an extra binner (-e l), use the
PKU-EMBL LorBin fork commit documented in the BASALT guide — not an
arbitrary upstream tip.
Prerequisites
| Requirement | Notes |
|---|
| Assembly | Long-read / hybrid contigs (metaFlye, metaMDBG, …) |
| Coverage | BAM or abundance table as required by upstream CLI |
| QC | Always follow with checkm2 (+ gunc for chimera risk) |
How to run
Verify flags against upstream README/--help (CLI evolves):
lorbin \
--contigs long_contigs.fa \
--bam temp/map/*.bam \
--outdir temp/bins/lorbin \
--threads 32
Decision tree
Long-read MAG recovery?
├─ Dedicated long-read binner → lorbin
├─ SemiBin2 long-read / ensemble-DBSCAN path → semibin2
├─ Multi-binner + refinement (related samples) → basalt ★
└─ Only short reads → prefer metabat2 / semibin2 / vamb (+ ensemble)
Methods caveats
- Unsupervised clustering + SCG gates ≠ experimental species proof.
- Report assembler, read type (ONT/HiFi), LorBin version, and CheckM2 gates.
- Do not claim “novel phylum” from LorBin alone — require GTDB-Tk + literature
placement and chimera screens (
gunc).
Related skills
metaflye · metamdbg · semibin2 · vamb · basalt · checkm2 · gunc ·
tool-selection · metagenomics-workflow
