| name | mmseqs2 |
| description | MMseqs2 — ultra-fast protein sequence search, clustering (Linclust), and contig taxonomy (Steinegger / Söding labs). Use when: (1) Building NR gene catalogues, (2) Sensitive BLASTP-scale search at metagenome size, (3) Contig taxonomy via easy-taxonomy / taxDB, (4) Contrasting with Foldseek structure search. Upstream: https://github.com/soedinglab/MMseqs2. For structure/motif search use foldseek / folddisco; route AI stack via metagenomics-llm.
|
| license | MIT |
| category | analysis-tools |
| tags | ["MMseqs2","cluster","search","taxonomy","Linclust"] |
| upstream | https://github.com/soedinglab/MMseqs2 |
| stage | function |
MMseqs2
Upstream: MMseqs2 · https://mmseqs.com
Citation
Steinegger, M. & Söding, J. MMseqs2 enables sensitive protein sequence searching
for the analysis of massive data sets. Nat. Biotechnol. 35, 1026–1028
(2017). https://doi.org/10.1038/nbt.3988
Mirdita, M. et al. Fast and sensitive taxonomic assignment to metagenomic
contigs. Bioinformatics 37, 3029–3031 (2021).
https://doi.org/10.1093/bioinformatics/btab184
See also docs/references.md.
Analytical thinking
MMseqs2 is the sequence workhorse of the Steinegger stack. Same design
philosophy as Foldseek (k-mer prefilter → ungapped → alignment), but on amino
acids rather than 3Di structure letters.
| Mode | Role in shotgun pipelines |
|---|
easy-cluster / Linclust | NR gene/protein catalogues (vs cd-hit) |
search / easy-search | Homology annotation at catalogue scale |
easy-taxonomy | Contig taxonomy: extract protein fragments → approx. 2bLCA → weighted vote |
Taxonomy mode covers all domains via fragment extraction (not prokaryote-only
ORF callers). Faster than CAT-style stacks on benchmarks in the 2021 note;
still a different claim from kraken2 / metabuli read profilers or
gtdbtk MAG taxonomy.
| vs | Prefer MMseqs2 when |
|---|
cd-hit | Large catalogues / speed |
diamond | Taxonomy workflows or Linclust-scale clustering |
foldseek | Sequence-only; no structures / ProstT5 |
kraken2 | Contig protein→tax labels (not k-mer read profile) |
How to run
mmseqs easy-cluster genes.faa temp/mmseqs/clu temp/mmseqs/tmp \
--min-seq-id 0.9 -c 0.8 --cov-mode 1
mmseqs easy-search query.faa targetDB aln.m8 temp/mmseqs/tmp
mmseqs easy-taxonomy contigs.fa taxonomyDB temp/mmseqs/tax temp/mmseqs/tmp
Key parameters
| Setting | Notes |
|---|
--min-seq-id / -c / --cov-mode | Catalogue identity & coverage — declare for NR |
-s | Sensitivity vs speed |
| taxDB | Required for easy-taxonomy; pin nr/UniProt/GTDB build |
| tmp dir | Fast local disk; large catalogues need RAM planning |
Decision tree
Sequence vs structure homology?
├─ Huge protein NR → mmseqs2 easy-cluster
├─ Classic small NR → cd-hit still fine
├─ Contig taxonomy (protein-based) → mmseqs2 easy-taxonomy
├─ BLASTP-like search → mmseqs2 / diamond
├─ Structure / twilight zone → foldseek (± ProstT5)
├─ Motif geometry → folddisco
└─ PLM remote homology, sequence-only → plmsearch / deepblast
Related skills
foldseek · folddisco · cd-hit · diamond · eggnog-mapper · prodigal ·
kraken2 · metabuli · metagenomics-llm · tool-selection
