sourmash sketch documentation

Most of the commands in sourmash work with signatures, which contain information about genomic or proteomic sequences. Each signature contains one or more sketches, which are compressed versions of these sequences. Using sourmash, you can search, compare, and analyze these sequences in various ways.

To create a signature with one or more sketches, you use the sourmash sketch command. There are three main commands:

sourmash sketch dna
sourmash sketch protein
sourmash sketch translate
sourmash sketch fromfile

The sketch dna command reads in DNA sequences and outputs DNA sketches.

The sketch protein command reads in protein sequences and outputs protein sketches.

The sketch translate command reads in DNA sequences, translates them in all six frames, and outputs protein sketches.

The sketch fromfile command takes in a CSV file containing the locations of genomes and proteomes, and outputs all of the requested sketches. It is primarily intended for large-scale database construction. (fromfile is a new command as of sourmash v4.4.0.)

All sourmash sketch commands take FASTA or FASTQ sequences as input; input data can be uncompressed, compressed with gzip, or compressed with bzip2. The output will be one or more signature files that can be used by other sourmash commands.

Quickstart

DNA sketches for genomes and reads

To create a DNA sketch for a genome, run:

sourmash sketch dna genome.fna

This will create an output file genome.fna.sig in the current directory, containing a single DNA signature for the entire genome, calculated using the default parameters.

Sourmash can work with unassembled reads; run

sourmash sketch dna -p k=21,k=31,k=51,abund metagenome.fq.gz

to create three abundance-weighted sketches at k=21, 31, and 51, for the given FASTQ file.

By default, sketch dna ignores bad k-mers (e.g. non-ACGT characters in DNA). If --check-sequence is provided, sketch dna will error exit on the first bad k-mer.

Building a combined sketch from two or more files

If you have multiple files, sourmash will by default create one sketch for each file. For situations such as paired-end read files from Illumina sequencing, you may instead want to build a combined sketch.

You can build a combined sketch in two ways.

First, you can use --name/--merge to build a single (named) sketch out of multiple input files:

sourmash sketch dna -p k=31 sample_R1.fq.gz sample_R2.fq.gz \
    --name "sample" -o sample.zip

Here you need to specify a name because sourmash does not pick a default name when given multiple files; you also need to provide an output file name because sourmash doesn’t pick a default output name in this situation.

Second, you can stream the input files into sourmash sketch via stdin:

gunzip -c sample_R?.fq.gz | sourmash sketch dna -p k=31 - \
    -o sample.zip

As above, you need to specify an output filename because sourmash can’t guess a good default for streaming input. The --name option can still be specified if you want to name the output sketch something other than -.

Note that the order of sequences or sequence files does not affect the output of sourmash sketch at all: you do not need to interleave reads or provide the input files in a consistent order.

Protein sketches for genomes and proteomes

The command:

sourmash sketch translate genome.fna

will output a protein sketch in ./genome.fna.sig, calculated by translating the genome sequence in all six frames and then using the default protein sketch parameters. K-mers may include stop codons and stop codons are considered valid protein-coding sequence.

And

sourmash sketch protein -p k=25,scaled=500 -p k=27,scaled=250 genome.faa

outputs two protein sketches to ./genome.faa.sig, one calculated with k=25 and scaled=500, the other calculated with k=27 and scaled=250.

If you want to use different encodings, you can specify them in a few ways; here is a parameter string that specifies a dayhoff encoding for the k-mers:

sourmash sketch protein -p k=25,scaled=500,dayhoff genome.faa

Translated DNA sketches for metagenomes

The command

sourmash sketch translate metagenome.fq

will take each read in the FASTQ file and translate the read into amino acid sequence in all six possible coding frames. No attempt is made to determine the right frame (but we are working on ways to determine this; see orpheum).

We suggest using this primarily on unassembled metagenome data. For most microbial genomes, it is both higher quality and more efficient to first predict the coding sequences (using e.g. prodigal) and then use sketch protein to build signatures.

Bulk sketch construction from many files

(This was added as of sourmash v4.4.0.)

The sourmash sketch fromfile command is intended for use when building many signatures as part of a larger workflow. It supports a variety of options to build new signatures, parallelize signature construction, and otherwise aid in tracking and managing database construction.

The command

sourmash sketch fromfile datasets.csv -p dna -p protein -o database.zip

will ingest a CSV spreadsheet containing (at a minimum) the three columns name, genome_filename, and protein_filename, and build all of the signatures requested by the parameter strings. Other columns in this file will be ignored.

If no protein, hp, or dayhoff sketches are requested, protein_filename can be empty for a given row; likewise, if no DNA sketches are requested, genome_filename can be empty for a given row.

Some of the key command-line options supported by fromfile are:

  • -o/--output-signatures will save generated signatures to any of the standard supported output formats.

  • -o/--output-csv-info will save a CSV file of input filenames and parameter strings for use with the sourmash sketch command line; this can be used to construct signatures in parallel.

  • --already-done will take a list of existing signatures/databases to check against; signatures with matching names and parameter strings will not be rebuilt.

  • --output-manifest-matching will output a manifest of already-existing signatures, which can then be used with sourmash sig cat to collate signatures across databases; see using manifests. (This provides sourmash sig check functionality in sketch fromfile.)

If you would like help and advice on constructing large databases, or pointers to code for generating the fromfile CSV format, please ask on the sourmash issue tracker or gitter support channel.

More detailed documentation

Input formats

sourmash sketch auto-detects and reads FASTQ or FASTA files, either uncompressed or compressed with gzip or bzip2. The filename doesn’t matter; sourmash sketch will figure out the format from the file contents.

You can also stream any of these formats into sourmash sketch via stdin by using - as the input filename. For example,

gunzip -c data/GCF*.fna.gz | sourmash sketch dna - -o out.sig

will make a single DNA signature from all of the FASTA sequences in data/GCF*.fna.gz.

Note, for signatures calculated from stdin, the signature filename attribute will be left empty, and sourmash sig describe will output ** no name **.

Input contents and output signatures

By default, sourmash sketch will produce signatures for each input file. If the file contains multiple FASTA/FASTQ records, these records will be merged into the output signature. You can provide a list of FASTA files in a text file to sourmash sketch by passing the text file path in via --from-file.

If you specify --singleton, sourmash sketch will produce signatures for each record.

If you specify --merge <name>, sourmash sketch will produce signatures for all input files and combine them into one signature.

The output signature(s) will be saved in locations that depend on your input parameters. By default, sourmash sketch will put the signatures in the current directory, in a file named for the input file with a .sig suffix. If you specify -o, all of the signatures will be placed in that file.

Protein encodings

sourmash sketch protein and sourmash sketch translate output protein sketches by default, but can also use the dayhoff and hp encodings. The Dayhoff encoding collapses multiple amino acids into a smaller alphabet so that amino acids that share biochemical properties map to the same character. The hp encoding divides amino acids into hydrophobic and polar (hydrophilic) amino acids, collapsing amino acids with hydrophobic side chains together and doing the same for polar amino acids.

We are still in the process of benchmarking these encodings; ask on the issue tracker if you are interested in updates.

Note that stop characters (*) are considered valid in all three encodings, and are not truncated. For example, amino acid sequences that contain stop characters at the end will produce a k-mer containing the stop character, and that k-mer will be hashed and potentially included in the sketch.

Parameter strings

The -p argument to sourmash sketch provides parameter strings to sourmash, and these control what signatures and sketches are calculated and output. Zero or more parameter strings can be given to sourmash. Each parameter string produces at least one sketch.

A parameter string is a space-delimited collection that can contain one or more fields, comma-separated.

  • k=<ksize> - create a sketch at this k-mer size; can provide more than one time in a parameter string. Typically ksize is between 4 and 100.

  • scaled=<int> - create a scaled MinHash with k-mers sampled deterministically at 1 per <scaled> value. This controls sketch compression rates and resolution; for example, a 5 Mbp genome sketched with a scaled of 1000 would yield approximately 5,000 k-mers. scaled is incompatible with num. See our guide to signature resolution for more information.

  • num=<int> - create a standard MinHash with no more than <num> k-mers kept. This will produce sketches identical to mash sketches. num is incompatible with scaled. See our guide to signature resolution for more information.

  • abund / noabund - create abundance-weighted (or not) sketches. See Classify signatures: Abundance Weighting for details of how this works.

  • dna, protein, dayhoff, hp - create this kind of sketch. Note that sourmash sketch dna -p protein and sourmash sketch protein -p dna are invalid; please use sourmash sketch translate for the former.

For all field names but k, if multiple fields in a parameter string are provided, the last one encountered overrides the previous values. For k, if multiple ksizes are specified in a single parameter string, sketches for all ksizes specified are created.

If a field isn’t specified, then the default value for that sketch type is used; so, for example, sourmash sketch dna -p abund would calculate a sketch with k=31,scaled=1000,abund. See below for the defaults.

Default parameters

The default parameters for sketches are as follows:

  • dna: k=31,scaled=1000,noabund

  • protein: k=10,scaled=200,noabund

  • dayhoff: k=16,scaled=200,noabund

  • hp: k=42,scaled=200,noabund

These were chosen by a committee of PhDs as being good defaults for an initial analysis, so, beware :).

More seriously, the DNA parameters were chosen based on the analyses done by Koslicki and Falush in MetaPalette: a k-mer Painting Approach for Metagenomic Taxonomic Profiling and Quantification of Novel Strain Variation.

The protein, dayhoff, and hp parameters were selected based on unpublished research results and/or magic formulas. We are working on publishing the results! Please ask on the issue tracker if you are curious.

More complex parameter string examples

Below are some more complicated sourmash sketch command lines:

  • sourmash sketch dna -p k=51 - default to a scaled=1000 and noabund for a k-mer size of 51 (based on moltype/command)

  • sourmash sketch dna -p k=31,k=51,k=21 - create one signature with multiple ksizes, using the defaults otherwise

  • sourmash sketch translate -p k=20,num=500,protein -p k=19,num=400,dayhoff,abund -p k=30,scaled=200,hp - create three signatures with different ksizes, moltypes, and scaled/num.

Signature naming

Signature names are displayed in the output for search, gather, and compare, and can be specified in a few different ways.

With default arguments, sourmash sketch does not set a name, and the filename is used in display output.

You can set a name using --name, but this has the side effect of merging the sequence records before signature creation. So, for example, sourmash sketch dna genome1.fa genome2.fa --name genome1 -o genome.sig would produce one signature after combining genome1.fa and genome2.fa.

The option --name-from-first will set the signature name from the first record header encountered in each file. When used with --singleton, this will name each signature based on the record that it is created from.

You can examine the signature name using sourmash sig describe.

Individual signature renaming can be done from the command line using sourmash sig split to create individual files for each signature, and then sourmash sig rename.

Locations for output files

Signature files can contain multiple signatures and sketches. Use sourmash sig fileinfo to summarize the contents of a signature file, and sourmash sig describe to get details on the contents of a file.

You can use -o <filename> to specify a file output location for all the output signatures; -o - means stdout. This does not merge signatures unless --merge is provided.

Specify --outdir to put all the signatures in a specific directory.

Output file formats

Sourmash can read and write signatures in many different formats, and sourmash sketch ... -o <filename> supports all of the standard output formats. Our recommendation is to output to zip files - e.g. filename.zip - as this is the smallest and most flexible signature storage format.

Please see Choosing signature output formats for more details.

Downsampling and flattening signatures

Creating signatures is probably the most time consuming part of using sourmash, and it is the only part that requires access to the raw data. Moreover, the output signatures are generally much smaller than the input data. So, we generally suggest creating a large set of signatures once.

To support this, sourmash can do two kinds of signature conversion without going back to the raw data.

First, you can downsample num and scaled signatures using sourmash sig downsample. For any sketch created with num parameter, you can decrease that num. And, for any scaled parameter, you can increase the scaled. This will decrease the size of the sketch accordingly; for example, going from a num of 5000 to a num of 1000 will decrease the sketch size by a factor of 5, and going from a scaled of 1000 to a scaled of 10000 will decrease the sketch size by a factor of 10.

(Note that decreasing num or increasing scaled will increase calculation speed and lower the accuracy of your results.)

Second, you can flatten abundances using sourmash sig flatten. For any sketch created with abund, you can convert it to a noabund sketch. This will decrease the sketch size, although not necessarily by a lot.

Unfortunately, changing the k-mer size or using different DNA/protein encodings cannot be done on a sketch, and you need to create new signatures from the raw data for that.

Examining the output of sourmash sketch

You can use sourmash sig describe to get detailed information about the contents of a signature file, and sourmash sig fileinfo to get a human-readable summary of the contents. This can help if you want to see exactly what a particular sourmash sketch command does!

Filing issues and asking for help

We try to provide good documentation and error messages, but may not succeed in answer all your questions! So we’re happy to help out!

Please post questions on the sourmash issue tracker. If you find something confusing or buggy about the documentation or about sourmash, we’d love to fix it – for you and for everyone else!