savont - Amplicon Sequence Variants (ASVs) and taxonomic profiling for long read amplicons

Savont generates Amplicon Sequence Variants (ASVs) at single-nucleotide resolution from long-read amplicon sequencing data such as

• Oxford Nanopore (ONT) R10.4 sequencing (preferably with SUP basecalling)
• PacBio HiFi sequencing

Savont differs from mapping-based approaches (e.g. Emu or ONT's epi2me workflow). Savont instead follows the Reads -> ASV -> Classification paradigm (just like DADA2, but for noisier long reads).

Why savont?

• Savont can resolve ASVs that differ by a single nucleotide, even for nanopore reads. This differs from existing nanopore workflows for OTU-like clustering.
• For ONT amplicons, savont requires ~10x less depth to generate ASVs compared to DADA2 / UNOISE.
• Savont also has built-in support for full taxonomic profiling (fastq -> abundance table -> QIIME) for several rRNA databases.

[!NOTE] Savont is optimized for long reads with >98% accuracy. ONT's R10.4 reads with SUP basecalling or PacBio HiFi are preferred. For lower quality reads (e.g. R9.4 ONT data) savont may not be useful.

<img width="90%" alt="github-diagram" src="https://github.com/user-attachments/assets/c0d9e356-ee1d-4d60-a217-c050e5abd0dc" />

Results

[!tip] The savont preprint is now out! Please see the preprint for results and methodological details.

ONT and HiFi results on 8-species community for full-length 16S sequences.

Diversity and ASV metrics on complex full-length ONT 16S amplicons.

Install via conda or build from source

Option 1: Build from source

Requirements: 1. rust (tested for > v1.88) programming language 2. Standard linux toolchain (tar, gzip, wget, C++, gcc) 3. cmake

git clone https://github.com/bluenote-1577/savont
cd savont

# Build and install
cargo install --path .
savont --help

Option 2: Conda

mamba install -c bioconda savont
# or use conda instead of mamba

Quick start for trimmed full-length 16S amplicons

Step 1: get ASVs from primer/adapter trimmed reads

You can either run savont in single sample mode or pool samples together. See documentation on multi-sample profiling.

# generate 16S ASVs for ONT / HiFi after cutadapt
savont asv 16s_full-length_trimmed.fastq.gz -o savont-out -t 20 (optional: --hifi)

# OR pool samples, get shared ASVs, quantify each sample independently
savont asv --pooled-samples sample1.fq.gz sample2.fq.gz  ...

Step 2: download databases and classify ASVs

# download Greengenes, EMU, or silva databases
savont download --location /path/databases --dbs greengenes2-2024.09 emu-1 silva-138.2

# classify against any database
savont classify -i savont-out -d /path/databases/emu-1 -t 20

Step 3: export results to QIIME2-parseable formats (optional)

# export a single run to QIIME2-compatible outputs
savont export -i savont-out -o export_output

# or combine multiple runs into one merged export
savont export -i savont-out1 savont-out2 -o export_output

Full subcommand guide

Generate ASVs from reads (savont asv)

[!NOTE] Savont filters reads based on length and quality. However, savont does not do adapter/primer trimming. Please trim your reads with e.g. cutadapt first.

Default savont parameters assume ONT / HiFi full-length 16S rRNA sequencing with trimmed reads from cutadapt. See below for parameter choices if this is not the case.

# Full-length 16S rRNA reads -> ASVs
savont asv 16s_full-length.fastq.gz -o savont-out -t 20 

# Full bacterial rRNA operon amplicons -> ASVs
savont asv operon_reads.fastq.gz -o savont-out -t 20 --rrna-operon

# For single-stranded protocols
savont asv 16s_single_strand.fq --single-strand -o savont-out -t 20

# Other types of amplicons with known length
savont asv amplicons.fastq.gz -o savont-out -t 20 --min-read-length 1600 --max-read-length 2100 

# Resulting ASVs
ls savont-out/final_asvs.fasta

Multi-sample savont workflow

You can either

1. Run savont on multiple samples independently and do savont export to dereplicate ASVs and merge (see below).

2. Or, you can pool samples through savont asv --pooled-samples sample1.fq sample2.fq .... This runs ASV detection on the concatenation of all samples, but sample-by-sample abundance quantification.

Pooling may allow recovery of low abundance ASVs and prevent cross-sample ASV splitting (see here). However, pooling is more computationally expensive, both in terms of memory and speed.

ASV output

The savont asv command produces:

1. final_asvs.fasta - Final ASV sequences (high-quality, chimera-filtered)
2. feature-table.tsv - QIIME2-compatible feature table (ASV × sample read counts)
3. final_clusters.tsv - Cluster assignments approximately mapping reads to ASVs
4. temp/ - Directory containing intermediate files

Taxonomic profiling against a reference database

Savont can also classify ASVs and generate a taxonomic profile with abundances. Savont supports two classification approaches:

• savont classify — minimap2 alignment against database with species- and genus-level output (better for species level). Uses identity thresholds for taxonomic assignment (thresholds from Yarza et al.)
• savont sintax — SINTAX k-mer bootstrap; genus-level only (better for unknown taxa)

Step 1: Download a reference database

# Download one or more databases (emu-1, silva-138.2, greengenes2-2024.09)
savont download --location /path/databases --dbs emu-1
savont download --location /path/databases --dbs silva-138.2
savont download --location /path/databases --dbs greengenes2-2024.09

# Or download multiple at once
savont download --location /path/databases --dbs emu-1 silva-138.2 

Step 2a: Classify ASVs with alignment (savont classify)

# Classify using any downloaded database (type is auto-detected)
savont classify -i savont-out -d databases/emu-1 -t 20
savont classify -i savont-out -d databases/silva-138.2 -t 20

# Adjust identity thresholds
savont classify -i savont-out -d databases/emu-1 \
    --species-threshold 99.9 --genus-threshold 90.0

Step 2b: Classify ASVs with SINTAX (savont sintax)

savont sintax uses 12-mer bootstrap resampling (100 iterations by default) to assign genus-level taxonomy with per-rank confidence scores.

savont sintax -i savont-out -d databases/emu-1 -t 20

# Adjust bootstrap threshold (default: 0.80)
savont sintax -i savont-out -d databases/silva-138.2 --min-bootstrap 0.70

Classification Output (savont classify)

The savont classify command produces three output files:

1. species_abundance.tsv / genus_abundance.tsv

Species-level or genus-level taxonomic abundance table:

abundance       species         genus   family  order   class   phylum  clade   superkingdom
0.45123         Escherichia_coli        Escherichia     Enterobacteriaceae      ...
0.23456         Staphylococcus_aureus   Staphylococcus  Staphylococcaceae       ...

• abundance - Relative abundance estimated by mappings and ASV depths
• Full taxonomic lineage from species to superkingdom

Note: *_abundance.tsv has extra per-sample abundances if --pooled-samples was specified during savont asv.

2. asv_mappings.tsv

Individual ASV mapping details:

asv_header      depth   alignment_identity  number_mismatches   tax_id  species             genus       family              order               class                   phylum          clade   superkingdom    reference
final_consensus_0_depth_5936    5936    99.67   5   29466   Veillonella parvula Veillonella Veillonellaceae Veillonellales  Negativicutes   Bacillota           Bacteria    29466:emu_db:36875
final_consensus_1_depth_3081    3081    99.27   11  29466   Veillonella parvula Veillonella Veillonellaceae Veillonellales  Negativicutes   Bacillota           Bacteria    29466:emu_db:36873

All taxonomic ranks from species to superkingdom are included. Lower taxonomic ranks may be UNCLASSIFIED if the alignment identity is low. Unmapped ASVs have UNCLASSIFIED in every rank column.

Classification Output (savont sintax)

savont sintax produces genus_abundance.tsv (same format as above) and asv_mappings.tsv with bootstrap confidence scores:

asv_header      depth   species_bootstrap   genus_bootstrap family_bootstrap    order_bootstrap class_bootstrap phylum_bootstrap    superkingdom_bootstrap  species         genus       family          ...
final_consensus_0_depth_5936    5936    0.000   0.980   0.990   0.995   0.999   1.000   1.000   UNCLASSIFIED    Veillonella Veillonellaceae  ...

Ranks below --min-bootstrap are reported as UNCLASSIFIED.

Export to QIIME2 format (savont export)

savont export converts savont output into QIIME2-importable files. It works on a single run directory or combines multiple directories into one merged profile.

ASVs across directories are matched by exact sequence hash. By default, savont also fuzzy-merges ASVs with identical sequences but trimmed to slightly different lengths across runs.

# Export a single run
savont export -i savont-out -o export-out

# Combine multiple runs
savont export -i sample1-out sample2-out -o export-out

# Combine many runs and give them meaningful names
savont export \
    -i run1/savont-out run2/savont-out run3/savont-out \
    -o export-out \
    --relabel SampleA SampleB SampleC

[!NOTE] Make sure to use --relabel if you have duplicate sample names.

Export outputs

1. merged_feature_table.tsv — QIIME2-compatible feature table; rows are hash-keyed ASVs, columns are samples
2. merged_rep_seqs.fasta — representative sequences for all merged ASVs
3. merged_asv_taxonomy.tsv — ASV-level taxonomy (Feature ID → lineage); only written if savont classify or savont sintax was run on any input directory
4. merged_taxon_counts.tsv — human-readable taxon count table; rows are lineage strings (Bacteria;Firmicutes;...), columns are sample counts; useful for quick inspection without QIIME2

Importing exported outputs into QIIME2 + stacked bar plot

# Feature table
biom convert -i export-out/merged_feature_table.tsv \
    -o feature-table.biom --table-type='OTU table' --to-hdf5
qiime tools import --type 'FeatureTable[Frequency]' \
    --input-path feature-table.biom --output-path feature-table.qza

# Representative sequences
qiime tools import --type 'FeatureData[Sequence]' \
    --input-path export-out/merged_rep_seqs.fasta --output-path rep-seqs.qza

# Taxonomy (if classify/sintax was run)
qiime tools import --type 'FeatureData[Taxonomy]' \
    --input-format HeaderlessTSVTaxonomyFormat \
    --input-path export-out/merged_asv_taxonomy.tsv --output-path taxonomy.qza

qiime taxa barplot --i-table feature-table.qza --i-taxonomy taxonomy.qza \
    --o-visualization taxa-bar-plots.qzv

Database Information

EMU (emu-1)

From Emu by Curry et al. (2022, Nature Methods). Curated 16S rRNA database with focused species-level classifications. Less breadth than SILVA or Greengenes2 for complex microbiomes.

SILVA (silva-138.2) — SSU Ref NR99 v138.2

More comprehensive than EMU, especially for understudied taxa. However, species-level classifications are often split across multiple distinct strains.

GreenGenes2 (greengenes2-2024.09)

GreenGenes2 2024.09 species-level trainset (DADA2 format). Unannotated ranks are reported as Greengenes_unannotated.

CHANGELOG

See the changelog.

Citation

Jim Shaw, Marie Riisgaard-Jensen, Kasper Skytte Andersen, Rasmus Kirkegaard, Morten Kam Dahl Dueholm, Heng Li. bioRxiv 2026.05.26.727271; doi: https://doi.org/10.64898/2026.05.26.727271

If you use any provided database, cite:

1. Quast, Christian, et al. "The SILVA ribosomal RNA gene database project: improved data processing and web-based tools." Nucleic acids research 41.D1 (2012): D590-D596.

2. Curry, Kristen D., et al. "Emu: species-level microbial community profiling of full-length 16S rRNA Oxford Nanopore sequencing data." Nature methods 19.7 (2022): 845-853.

3. McDonald, D., Jiang, Y., Balaban, M. et al. Greengenes2 unifies microbial data in a single reference tree. Nat Biotechnol 42, 715–718 (2024).

If you use the SINTAX algorithm please cite: Edgar, Robert C. "SINTAX: a simple non-Bayesian taxonomy classifier for 16S and ITS sequences." biorxiv (2016): 074161.

License

MIT