Metabuli classifies metagenomic reads by comparing them to reference genomes. You can use Metabuli to profile the taxonomic composition of your samples or to detect specific (pathogenic) species.
Sensitive and Specific. Metabuli uses a novel k-mer structure, called metamer, to analyze both amino acid (AA) and DNA sequences. It leverages AA conservation for sensitive homology detection and DNA mutations for specific differentiation between closely related taxa.
A laptop is enough. Metabuli operates within user-specified RAM limits, allowing it to search any database that fits in storage. A PC with 8 GiB of RAM is sufficient for most analyses.
A few clicks are enough. Metabuli App is now available here. With just a few clicks, you can run Metabuli and browse the results with Sankey and Krona plots on your PC.
Short reads, long reads, and contigs. Metabuli can classify all types of sequences.
For more details, please see Nature Methods, PDF, bioRxiv, or ISMB 2023 talk.
🖥️ Metabuli App for Windows, MacOS, and Linux are now available!
Run taxonomic profiling in just a few clicks and explore results with Sankey and Krona plots.
Download the app for your OS here—no separate Metabuli installation needed.
- Fix errors in v1.0.9
- Custom DB creation became easier
- Improve
updateDBcommand
- DB creation process improved
updateDBmodule to add new sequences to an existing database.- Users can provide CDS information to skip Prodigal's gene prediction.
--max-ramparameter added tobuildmodule.- Compatibility with taxdump files generated using taxonkit.
- Please check release note for details.
- Added
extractmodule to extract reads classified into a certain taxon.
- Metabuli became faster 🚀
- Windows: 8.3 times faster
- MacOS: 1.7 times faster
- Linux: 1.3 times faster
- Test details are in release note.
- Fixed a bug in score calculation that could affect classification results.
- Installation
- Pre-built databases
- Classification
- Refine Classification file
- Extract
- GTDB-based custom database
- NCBI or custom taxonomy based database
- Example
# install via conda
conda install -c conda-forge -c bioconda metabuli
# Linux AVX2 build (fast, recommended for most Linux system
# check using: cat /proc/cpuinfo | grep avx2)
wget https://mmseqs.com/metabuli/metabuli-linux-avx2.tar.gz; tar xvzf metabuli-linux-avx2.tar.gz; export PATH=$(pwd)/metabuli/bin/:$PATH
# Linux SSE2 build (slower, for old systems)
wget https://mmseqs.com/metabuli/metabuli-linux-sse2.tar.gz; tar xvzf metabuli-linux-sse2.tar.gz; export PATH=$(pwd)/metabuli/bin/:$PATH
# MacOS (Universal, works on Apple Silicon and Intel Macs)
wget https://mmseqs.com/metabuli/metabuli-osx-universal.tar.gz; tar xvzf metabuli-osx-universal.tar.gz; export PATH=$(pwd)/metabuli/bin/:$PATH
Metabuli also works on Linux ARM64 and Windows systems. Please check https://mmseqs.com/metabuli for static builds for other architectures.
git clone --recurse-submodules https://github.com/steineggerlab/Metabuli.git
cd Metabuli
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make -j 16
The built binary can be found in ./build/src.
You can download pre-built databases using databases workflow.
NOTE: The databases workflow may not work if you don't use the latest version of Metabuli.
In that case, please manually download databases from this link.
Usage:
metabuli databases DB_NAME OUTDIR tmp
# NOTE
- A human genome (T2T-CHM13v2.0) is included in all databases below.
1. RefSeq Virus (8.1 GiB)
- NCBI RefSeq release 223 virus genomes
- Database will be in OUT_DIR/refseq_virus
metabuli databases RefSeq_virus OUT_DIR tmp
2. RefSeq Prokaryote and Virus (115.6 GiB)
- RefSeq prokaryote genomes (Complete Genome/Chromosome, 2024-03-26) + RefSeq Virus above.
- Database will be in OUT_DIR/refseq_prokaryote_virus
metabuli databases RefSeq OUTDIR tmp
3. GTDB (101 GiB)
- GTDB 214.1 (Complete Genome/Chromosome, CheckM completeness > 90 and contamination < 5).
- Database will be in OUT_DIR/gtdb
metabuli databases GTDB OUTDIR tmp
4. RefSeq Releases 224 (619 GiB)
- Viral and prokaryotic genomes of RefSeq release 224.
metabuli databases RefSeq_release OUTDIR tmp
Downloaded files are stored in OUTDIR/DB_NAME directory, which can be provided for classify module as DBDIR.
metabuli classify <i:FASTA/Q> <i:DBDIR> <o:OUTDIR> <Job ID> [options]
- INPUT : FASTA/Q file of reads you want to classify. (gzip supported)
- DBDIR : The directory of reference DB.
- OUTDIR : The directory where the result files will be generated.
- Job ID: It will be the prefix of result files.
# Paired-end
metabuli classify read_1.fna read_2.fna dbdir outdir jobid
# Single-end
metabuli classify --seq-mode 1 read.fna dbdir outdir jobid
# Long-read
metabuli classify --seq-mode 3 read.fna dbdir outdir jobid
* Important parameters:
--threads : The number of threads used (all by default)
--max-ram : The maximum RAM usage. (128 GiB by default)
--min-score : The minimum score to be classified
--min-sp-score : The minimum score to be classified at or below species rank.
--taxonomy-path: Directory where the taxonomy dump files are stored. (DBDIR/taxonomy by default)
--accession-level : Set 1 to use accession level classification (0 by default).
It is available when the DB is also built with accession level taxonomy.
- Paratemers for precision mode (Metabuli-P)
- Illumina short reads:
--min-score 0.15 --min-sp-score 0.5 - PacBio HiFi reads:
--min-score 0.07 --min-sp-score 0.3 - PacBio Sequel II reads:
--min-score 0.005 - ONT reads:
--min-score 0.008
- Illumina short reads:
This will generate three result files: JobID_classifications.tsv, JobID_report.tsv, and JobID_krona.html.
Sankey diagram is available in the GUI app.
You can use --lineage 1 option in classify module to print the full lineage next to rank column.
is_classified: Classified or notname: Read IDtaxID: Tax. ID in the tax. dump files used in database creationquery_length: Effective read lengthscore: DNA level identity scorerank: Taxonomic rank of the taxontaxID:match_count: List of "taxID : k-mer match count"
#is_classified name taxID query_length score rank taxID:match_count
1 read_1 2688 294 0.627551 subspecies 2688:65
1 read_2 2688 294 0.816327 subspecies 2688:78
0 read_3 0 294 0 no rank
It follows Kraken2's report format. The first line is a header, and the rest of the lines are tab-separated values. The columns are as follows:
clade_proportion: Percentage of reads classified to the clade rooted at this taxonclade_count: Number of reads classified to the clade rooted at this taxontaxon_count: Number of reads classified directly to this taxonrank: Taxonomic rank of the taxontaxID: Tax ID according to the taxonomy dump files used in the database creationname: Taxonomic name of the taxon
#clade_proportion clade_count taxon_count rank taxID name
33.73 77571 77571 no rank 0 unclassified
66.27 152429 132 no rank 1 root
64.05 147319 2021 superkingdom 8034 d__Bacteria
22.22 51102 3 phylum 22784 p__Firmicutes
22.07 50752 361 class 22785 c__Bacilli
17.12 39382 57 order 123658 o__Bacillales
15.81 36359 3 family 126766 f__Bacillaceae
15.79 36312 26613 genus 126767 g__Bacillus
2.47 5677 4115 species 170517 s__Bacillus amyloliquefaciens
0.38 883 883 subspecies 170531 RS_GCF_001705195.1
0.16 360 360 subspecies 170523 RS_GCF_003868675.1
It is for an interactive taxonomy report (Krona). You can use any modern web browser to open JobID_krona.html.
Metabuli can classify reads against a database of any size as long as the database is fits in the hard disk, regardless of the machine's RAM size. We tested it with a MacBook Air (2020, M1, 8 GiB), where we classified about 15 M paired-end 150 bp reads (~5 GiB in size) against a database built with ~23K prokaryotic genomes (~69 GiB in size).
After running the 'classify' command, you can refine classification file with some options.
metabuli classifiedRefiner <i:read-by-read classification> <i:DBDIR> [options]
- read-by-read classification : The JobID_classifications.tsv file generated by the `classify` step.
- DBDIR : The same DBDIR used in the `classify` step.
* Options
--threads : The number of threads to utilize (all by default)
--remove-unclassified : Remove unclassified reads
--exclude-taxid : Remove list of taxids as well as its children (e.g., 1758,9685,1234)
--select-taxid : Select list of taxids as well as its children (e.g., 1758,9685,1234)
--select-columns : Select list of columns with number and handle full lineage as 7 (generated if absent) (e.g., 2,5,7,3)
--report : Write report of refined classification file
--rank : Adjust classification to the specified rank
--rank-file-type : Choose how to handle reads assigned to higher taxonomic ranks when using the --rank option. [0: exclude higher rank, 1: include higher rank, 2: make separate file for higher rank classification]
- refined classification file :
JobID_refined.tsv - report :
JobID_refined_report.tsv,JobID_refined_krona.html - higher rank classification file :
_refined_higherRanks.tsv
After running the classify command, you can extract reads that are classified under a specific taxon.
This requires the FASTA/Q files used in the classify step and the JobID_classifications.tsv file, which is generated as one of the output files.
metabuli extract <i:FASTA/Q> <i:read-by-read classification> <i:DBDIR> --tax-id TAX_ID
- FASTA/Q : The FASTA/Q file(s) used during the `classify` step.
- read-by-read classification : The JobID_classifications.tsv file generated by the `classify` step.
- DBDIR : The same DBDIR used in the `classify` step.
- TAX_ID : The taxonomy ID of the taxon at any rank (e.g., species, genus) from which you want to extract the reads.
# Paired-end
metabuli extract read_1.fna read_2.fna JobID_classifications.tsv dbdir --tax-id TAX_ID
# Single-end
metabuli extract --seq-mode 1 read.fna JobID_classifications.tsv dbdir --tax-id TAX_ID
# Long-read
metabuli extract --seq-mode 3 read.fna JobID_classifications.tsv dbdir --tax-id TAX_ID
- For paired-end samples:
read_1_TAX_ID.fnaandread_2_TAX_ID.fna - For single-end or long-read samples:
read_TAX_ID.fna
User-provided CDS information (optional): Use
--cds-infoto provide absolute paths to CDS files. For included accessions, the provided CDS is used, and Prodigal's gene prediction is skipped. Only GenBank/RefSeq CDS files are supported (e.g., GCA_000839185.1_ViralProj14174_cds_from_genomic.fna.gz).
Important
Requirements: Reference FASTA file name (or path) must include the assembly accession (e.g., GCF_028750015.1, regexGC[AF]_[0-9]+\.[0-9]+). Files from RefSeq or GenBank meet this requirement.
1. Download taxonkit-generated GTDB taxdump files here.
# GTDB_TAXDUMP: Directory with downloaded GTDB taxdump files.
# FASTA_LIST: File of reference genome absolute paths.
# DBDIR: Directory where the database will be generated.
metabuli build --gtdb 1 <DBDIR> <FASTA_LIST> <GTDB_TAXDUMP/taxid.map> --taxonomy-path <GTDB_TAXDUMP> [options]
* Options
--threads : The number of threads to utilize (all by default)
--max-ram : The maximum RAM usage. (128 GiB by default)
--accession-level : Set 1 to creat a DB for accession level classification (0 by default).
--cds-info : List of absolute paths to CDS files.
This will generate diffIdx, info, split, and taxID_list and some other files. You can delete *_diffIdx and *_info files.
Note
If you want to use new GTDB release, please build a new database from scratch.
You can add new sequences to a GTDB-based database. Expanding the taxonomy for virus or eukaryote is also possible.
# GTDB_TAXDUMP: Directory with downloaded GTDB taxdump files.
# FASTA_LIST: File of absolute paths to new sequences.
# NEW DBDIR: Updated database is generated here.
# OLD DBDIR: Directory of an existing database.
metabuli updateDB --gtdb 1 <NEW DBDIR> <FASTA_LIST> <GTDB_TAXDUMP/taxid.map> <OLD DBDIR> [options]
* Options
--make-library: When many species are in the same FASTA, enable it for faster execution (0 by default).
--new-taxa: List of new taxa to be added.
--threads: The number of threads to utilize (all by default)
--max-ram: The maximum RAM usage. (128 GiB by default)
--accession-level: Set 1 to add new sequences for accession level classification (0 by default).
--cds-info: List of absolute paths to CDS files.
Warning
Mixing taxonomies within the same domain is not recommended. For example, adding prokaryotes to a GTDB database using NCBI taxonomy will cause issues, but adding eukaryotes or viruses using NCBI taxonomy is fine since GTDB does not cover them.
1. Check taxdump files to see if you need to add new taxa. taxdump command retrieves taxdump files of an existing database.
2-1. Create a new taxa list
If you have accession2taxid and taxonomy dump files of the new sequences, you can use createnewtaxalist to create an input for --new-taxa option. If not, you have to prepare the input manually (see below).
metabuli createnewtaxalist <OLD DBDIR> <FASTA_LIST> <new taxonomy dump> <accession2taxid> <OUTDIR>
It generates newtaxa.tsv for --new-taxa option and newtaxa.accession2taxid.
Suppose you're adding eukaryotes to a GTDB database. As GTDB doesn't include eukaryotes, you may want to use NCBI taxonomy for eukaryotes.
You can download taxdump files from here and accession2taxid from here.
metabuli createnewtaxalist <GTDB dir> <new seq list> <NCBI taxdump dir> <NCBI accession2taxid> <out dir>
metabuli updateDB <new db dir> <new seq list> <out dir/newtaxa.accession2taxid> <GTDB dir> --new-taxa <out dir/newtaxa.tsv>
2-2. Manually prepare a new taxa list
For the --new-tax option, provide a four-column TSV file in the following format.
taxID parentID rank name
The new taxon must be linked to a taxon in the existing database's taxonomy.
Suppose you want to add Saccharomyces cerevisiae to a GTDB database.
After inspecting taxonomy with taxdump, you find that the taxonomy lacks the Fungi kingdom and only includes one eukaryote (Homo sapiens). In this scenario, your new taxa list and accession2taxid should be as follows.
# New taxa list
## taxid parentTaxID rank name // Don't put this header in your actual file.
10000013 10000012 species Saccharomyces cerevisiae
10000012 10000011 genus Saccharomyces
10000011 10000010 family Saccharomycetaceae
10000010 10000009 order Saccharomycetales
10000009 10000008 class Saccharomycetes
10000008 10000007 phylum Ascomycota
10000007 10000000 kingdom Fungi // 10000000 is Eukaroyte taxID of the pre-built DB.
# accession2taxid
accession accession.version taxid gi
newseq1 newseq1 10000013 0
newseq2 newseq2 10000013 0
Important
Three requirements:
- FASTA files : Each sequence must have a unique
>accession.versionor>accesionheader (e.g.,>CP001849.1or>CP001849). - NCBI-style accession2taxid : Sequences with accessions absent here are skipped, and versions are ignored.
- NCBI-style taxonomy dump :
names.dmp,nodes.dmp, andmerged.dmp. Sequences with tax. IDs absent here are skipped.
- Download
accession2taxidfrom here. - Download
taxdumpfiles from here. - For custom sequences, edit
accession2taxidandtaxdumpfiles as follows.accession2taxid- For a sequence whose header is
>custom, addcustom[tab]custom[tab]taxid[tab]anynumber. - As above, version number is not necessary.
taxidmust be included in thenodes.dmpandnames.dmp.- Put any number for the last column. It is not used in Metabuli.
- For a sequence whose header is
taxdump- Edit
nodes.dmpandnames.dmpif you introduced a newtaxidinaccession2taxid.
- Edit
# DBDIR: Directory where the database will be generated.
# FASTA_LIST: A file containing absolute paths to FASTA files.
# accession2taxid : NCBI-style accession2taxid file.
# TAXDUMP: Directory with taxonomy dump files.
metabuli build <DBDIR> <FASTA_LIST> <accession2taxid> --taxonomy-path <TAXDUMP> [options]
* Options
--make-library: When many species are in the same FASTA, enable it for faster execution (0 by default).
--threads: The number of threads to use (all by default)
--max-ram: The maximum RAM usage. (128 GiB by default)
--accession-level: Set 1 to creat a DB for accession level classification (0 by default).
--cds-info: List of absolute paths to CDS files.
This will generate diffIdx, info, split, and taxID_list and some other files. You can delete *_diffIdx and *_info files and DATE-TIME folder (e.g., 2025-1-24-10-32) if generated.
You can add new sequences to an existing database, of which taxonomy will be used. You can add new taxa if the previous taxonomy does not include them (see "Add sequences of new taxa" below).
1. Prepare two files
- New FASTA file list : Each sequence must have a unique
>accession.versionor>accesionheader. - NCBI-style accession2taxid : Sequences with accessions absent here are skipped. Put any number in the GI column. Version number is ignored.
accession accession.version taxID gi SequenceA SequenceA.1 960611 0 SequenceB SequenceB.1 960612 0 NoVersionOkay NoVersionOkay 960613 0
2. Update database
# NEW DBDIR: Directory where the updated database will be generated.
# FASTA_LIST: A file of paths to new FASTA files.
# accession2taxid : NCBI-style accession2taxid file.
# OLD DBDIR: Directory of an existing database.
metabuli updateDB <NEW DBDIR> <FASTA_LIST> <accession2taxid> <OLD DBDIR> [options]
- FASTA list: A file of paths to the FASTA file to be added.
- accession2taxid : A path to NCBI-style accession2taxid.
* Options
--threads : The number of threads used (all by default)
--max-ram : The maximum RAM usage. (128 GiB by default)
--accession-level : Set 1 to create a DB for accession level classification (0 by default).
--make-library : Make species library for faster execution (1 by default).
--new-taxa : List of new taxa to be added.
<Add sequences of new taxa> - Please refer this section.
The example here was detecting SARS-CoV-2 variant-specific reads, but has changed since the pre-built DB no longer contains the variant genomes.
Classifying RNA-seq reads from a COVID-19 patient. The whole process must take less than 10 mins using a personal machine.
metabuli databases RefSeq_virus OUTDIR tmp
fasterq-dump --split-files SRR14484345
Download SRA Toolkit containing
fasterq-dumphere
metabuli classify SRR14484345_1.fq SRR14484345_2.fq OUTDIR/refseq_virus RESULT_DIR JOB_ID --max-ram RAM_SIZE
Find a section like the example below
92.2331 510490 442 species 694009 Severe acute respiratory syndrome-related coronavirus
92.1433 509993 509993 no rank 2697049 Severe acute respiratory syndrome coronavirus 2
Shen, W., Ren, H., TaxonKit: a practical and efficient NCBI Taxonomy toolkit, Journal of Genetics and Genomics, https://doi.org/10.1016/j.jgg.2021.03.006