---

Fast extraction and origin-sorting of small-subunit rRNA (16S/18S) from environmental sequencing reads.

SSUplex finds small-subunit rRNA in sequencing reads, assigns each read to one of five origins (bacteria, archaea, eukaryota, mitochondria, chloroplast), clips it to the SSU region and writes one FASTA file per origin, so that each set can be passed to an appropriate taxonomic classifier. It reimplements the HMM-based SSU extraction and origin-assignment step of Metaxa2 in Rust, using Metaxa2's own SSU profiles. It does not perform taxonomic classification.

Why sort reads by origin

Broadly conserved rRNA primers amplify more than the intended target: a 16S run from plant or host material can contain many chloroplast and mitochondrial reads, and 18S or cross-domain reads also appear. Sent directly to a bacterial 16S classifier, chloroplast reads tend to be assigned to Cyanobacteria, mitochondrial reads to Rickettsiales, and eukaryotic reads to low-confidence hits. SSUplex separates these reads first. For each read it:

1. detects SSU rRNA with profile HMMs of the gene's conserved regions, searching both strands with nhmmer, so reads in either orientation are found without a separate orientation step;
2. assigns an origin from the per-region scores of each origin's profiles;
3. extracts the SSU part of the read, clipped to the envelope of its hits.

Installation

Bioconda (recommended)

conda create -n ssuplex -c conda-forge -c bioconda ssuplex
conda activate ssuplex
ssuplex --version

This also installs HMMER, which provides nhmmer, the only runtime dependency.

From source

You need Rust 1.85 or newer (install with rustup) and HMMER 3.1 or newer with nhmmer on your PATH (for example conda install -c bioconda hmmer or brew install hmmer).

git clone https://github.com/ayobi/ssuplex.git
cd ssuplex
cargo install --locked --path .
ssuplex --version

cargo install builds a release binary and copies it to ~/.cargo/bin, which rustup adds to your PATH. --locked builds with exactly the dependency versions recorded in Cargo.lock. If your shell reports ssuplex: command not found, either add ~/.cargo/bin to PATH or build in place with cargo build --release --locked and run ./target/release/ssuplex.

HMM profiles

SSUplex needs one profile file per origin (bacteria.hmm, archaea.hmm, eukaryota.hmm, mitochondria.hmm, chloroplast.hmm) in the directory given by --hmm-dir (default ./hmms). They are built from the Metaxa2 SSU database, which is distributed with Metaxa2 under its own license, so they are not bundled. A working Metaxa2 installation is not required, only its download:

curl -L -O https://microbiology.se/sw/Metaxa2_2.2.3.tar.gz
tar xzf Metaxa2_2.2.3.tar.gz
dev/assemble_metaxa2_hmms.sh Metaxa2_2.2.3/metaxa2_db hmms

The script concatenates Metaxa2's per-origin profile files (A archaea, B bacteria, C chloroplast, E eukaryota, and M plus N for mitochondria, where N is Metaxa2's metazoan mitochondrial set) and prints how many profiles each file contains. The equivalent commands are:

SRC=Metaxa2_2.2.3/metaxa2_db/SSU/HMMs
mkdir -p hmms
cat "$SRC/B.hmm"              > hmms/bacteria.hmm
cat "$SRC/A.hmm"              > hmms/archaea.hmm
cat "$SRC/E.hmm"              > hmms/eukaryota.hmm
cat "$SRC/M.hmm" "$SRC/N.hmm" > hmms/mitochondria.hmm
cat "$SRC/C.hmm"              > hmms/chloroplast.hmm

Quick start

The Metaxa2 download includes test.fasta, 100 sequences with known content: 50 SSU sequences (ten from each origin) and 50 LSU sequences, which SSUplex should leave unclassified. From the directory where you assembled hmms/:

ssuplex -i Metaxa2_2.2.3/test.fasta -o quickstart/test --hmm-dir hmms --threads 4
cat quickstart/test.summary.txt

The log should end with classified 50 of 100 read(s) to an origin. With HMMER 3.4 and default settings, the 50 SSU sequences are assigned 12 to bacteria, 10 each to archaea and eukaryota, and 9 each to mitochondria and chloroplast: one mitochondrial and one chloroplast sequence end up as bacteria (see Origin ranking). Running again with the same -o prefix stops with an error instead of overwriting the results; add --force to replace them.

Output files

For -o results/sample01:

File	Contents
sample01.bacteria.fasta	Reads assigned to bacteria, clipped to the SSU region
sample01.archaea.fasta	Reads assigned to archaea
sample01.eukaryota.fasta	Reads assigned to eukaryota (nuclear 18S)
sample01.mitochondria.fasta	Reads assigned to mitochondria
sample01.chloroplast.fasta	Reads assigned to chloroplast
sample01.unclassified.fasta	Reads without a qualifying SSU hit, unclipped
sample01.extraction.tsv	Per read: origin, score, region count, E-value, coordinates
sample01.summary.txt	Run settings and counts per origin
sample01.scores.tsv	With --debug-scores: per read and origin, region count, mean and summed bit score
sample01.regions.tsv	With --debug-scores: every per-region hit

A per-origin FASTA file is written only when at least one read is assigned to that origin.

Options

$ ssuplex -h
Fast Rust preprocessor for rRNA-marker eDNA workflows: extracts SSU rRNA from environmental sequencing reads and sorts them by origin (bacterial, archaeal, eukaryotic, mitochondrial, chloroplast).

Usage: ssuplex [OPTIONS] --input <FASTA> --output <PREFIX>

Options:
  -i, --input <FASTA>              Input FASTA of sequencing reads or contigs (optionally gzip-compressed)
  -o, --output <PREFIX>            Output prefix. Files will be written as `{prefix}.{origin}.fasta`, `{prefix}.extraction.tsv`, and `{prefix}.summary.txt`
      --hmm-dir <DIR>              Directory containing origin-specific HMM profiles (`bacteria.hmm`, `archaea.hmm`, `eukaryota.hmm`, `mitochondria.hmm`, `chloroplast.hmm`) [default: hmms]
  -t, --threads <THREADS>          Number of worker threads [default: number of available CPUs]
  -E, --evalue <EVALUE>            HMMER E-value cutoff for `nhmmer` [default: 0.00001]
      --min-score <MIN_SCORE>      Minimum mean per-region bit score the chosen origin must reach; otherwise the read is reported as `unclassified` [default: 0]
      --min-domains <MIN_DOMAINS>  Minimum number of conserved-region profiles that must match the chosen origin; otherwise the read is reported as `unclassified`. (Metaxa2's default is 2, accepting single-region hits only at E-value 1e-10 or lower.) [default: 1]
      --rank <RANK>                Statistic used to rank candidate origins for each read: `sum` (summed per-region bit score, the default), `mean` (mean per-region bit score), or `count` (number of matched regions). See the README section "Origin ranking" [default: sum] [possible values: mean, sum, count]
      --chunk-size <CHUNK_SIZE>    Number of sequences per HMMER batch. Larger batches reduce subprocess overhead; smaller batches improve memory locality and progress reporting granularity [default: 10000]
      --no-clip                    Keep extracted reads as full input sequences (no clipping). By default, SSUplex clips each read to the SSU coordinate envelope from HMM hits
      --debug-scores               Also write `{prefix}.scores.tsv` (per read and matched origin: n_regions, mean, sum, best_evalue) and `{prefix}.regions.tsv` (every per-region hit with score and coordinates), for diagnostics
      --force                      Replace output files left by a previous run with the same prefix. Without this flag SSUplex stops instead of overwriting existing results
  -v, --verbose...                 Increase logging verbosity. Repeat (`-vv`) for trace output
  -h, --help                       Print help (see more with '--help')
  -V, --version                    Print version

Origin ranking

For each read and origin, SSUplex keeps at most one hit per conserved-region profile and summarises the hits as a region count, a summed bit score and a mean bit score. --rank chooses which of these decides the origin:

--rank	Origin chosen by	Ties broken by
sum (default)	highest summed bit score	number of regions
mean	highest mean per-region bit score	number of regions
count	most matched regions	mean bit score

These mirror Metaxa2's --selection_priority options sum, score (Metaxa2's default) and domains, with one difference: Metaxa2 divides its sum by the number of profiles in the origin's set, and SSUplex does not. Because mitochondria.hmm combines Metaxa2's M and N sets, a mitochondrial origin can collect hits from both. A read is reported as unclassified when the chosen origin matches fewer than --min-domains regions or its mean score is below --min-score.

Why sum is the default. On noisy long reads, the true origin of a read typically matches more conserved regions than a related origin does, but each match scores a little lower. The mean ignores how many regions matched: on the ZymoBIOMICS Nanopore mock community it assigned more than half of the bacterial reads to chloroplast or mitochondria, and so does Metaxa2's own HMM-based step, which also ranks by the mean. The sum takes breadth into account and was the only statistic that stayed accurate on both clean reference sequences and noisy Nanopore reads (dev/BENCHMARK.md).

Its weakness is the opposite case: organellar reads that also match many bacterial profiles can be assigned to bacteria. In the benchmarks this affected about one in five full-length chloroplast reference sequences, more than a third of the reads that Metaxa2 (with BLAST) called mitochondrial in a plant-root sample, and 22 to 38% of simulated mitochondrial and 8 to 12% of simulated chloroplast reads. Metaxa2 resolves such reads with its BLAST step; with SSUplex, expect some organellar reads in the bacterial output. Classifiers whose reference databases include organelle sequences, such as SILVA, will label them as chloroplast or mitochondria. Where the organellar fraction itself is what you are measuring, --rank mean recovers more organellar reads from accurate long reads (98.5% identity or better), at the cost of sending many bacterial reads to organelles.

Relation to Metaxa2

SSUplex uses Metaxa2's SSU profiles and reproduces its extraction and HMM-based origin assignment. Metaxa2's default pipeline then runs a BLAST classification against its reference database, and its final origin calls also take those BLAST matches into account; metaxa2 -x T runs the extraction step alone. SSUplex has no classification step and leaves taxonomy to a dedicated classifier. Other differences: SSUplex searches with nhmmer (Metaxa2 uses hmmsearch on both strands), ranks origins by summed rather than mean score by default, uses an E-value cutoff of 1e-5 and one matching region by default (Metaxa2 requires two domains, or one domain at E-value 1e-10 or lower), and scores Metaxa2's M and N mitochondrial sets together.

Scope and limitations

• SSUplex is built and benchmarked as a preprocessing step for full-length 16S/18S amplicons sequenced with long reads (Nanopore, and reads at PacBio HiFi accuracy). It will run on short reads and on shotgun reads, but neither use has been benchmarked: a short read spans only one or two of the conserved regions that origin assignment relies on. The whole input is held in memory, so peak memory grows with input size.
• Organellar reads that also match bacterial profiles are the least certain calls; see Origin ranking.
• The profiles come from the Metaxa2 database (built from SILVA release 111 and Mitozoa release 10), so lineages that are poorly represented there may be detected less reliably.

Reproducing the benchmarks

dev/BENCHMARK.md describes the datasets, the exact commands, and how accuracy, run time and memory are measured. dev/smoke.sh runs an end-to-end test on synthetic profiles and reads.

Citation

O'Brien A, Vargas Botia J, Acuña I, Restovic F, Martínez P, Parada P. SSUplex: fast, both-strand extraction and origin-sorting of small-subunit rRNA for environmental DNA metabarcoding. bioRxiv (2026). doi: 10.64898/2026.07.02.736232

Version 0.2.0 of the software is archived on Zenodo: doi: 10.5281/zenodo.23246210

Please also cite Metaxa2, whose profiles and extraction approach SSUplex uses:

Bengtsson-Palme J, Hartmann M, Eriksson KM, Pal C, Thorell K, Larsson DGJ, Nilsson RH (2015). Metaxa2: Improved identification and taxonomic classification of small and large subunit rRNA in metagenomic data. Molecular Ecology Resources 15(6):1403-1414. doi: 10.1111/1755-0998.12399

License

MIT; see LICENSE. The HMM profiles you build are derived from the Metaxa2 database and remain under Metaxa2's license.