Split k-mer analysis (version 2) uses exact matching of split k-mer sequences to align closely related sequences, typically small haploid genomes such as bacteria and viruses. Recall drops beyond around 1% divergence (see paper).

SKA can only align SNPs further than the k-mer length apart, and does not use a gap penalty approach or give alignment scores. But the advantages are speed and flexibility, particularly the ability to run on a reference-free manner (i.e. including accessory genome variation) on both assemblies and reads.

ska.rust also incorporates 'ska lo' (left-out), which uses local assembly to recover more variants in more diverged samples, and can be run as well as/instead of ska align.

ska:

Romain Derelle, Johanna von Wachsmann, Tommi Mäklin, Joel Hellewell, Timothy Russell, Ajit Lalvani, Leonid Chindelevitch, Nicholas J. Croucher, Simon R. Harris, John A. Lees (2024). Seamless, rapid, and accurate analyses of outbreak genomic data using split k-mer analysis. Genome Research, 34(10), 1661–1673.

https://genome.cshlp.org/content/34/10/1661.abstract

ska lo:

Romain Derelle, Kieran Madon, Joel Hellewell, Víctor Rodríguez-Bouza, Nimalan Arinaminpathy, Ajit Lalvani, Nicholas Croucher, Simon Harris, John Lees, Leonid Chindelevitch (2025). Reference-free variant calling with local graph construction with ska lo (SKA). Molecular Biology and Evolution, msaf077.

https://academic.oup.com/mbe/article/42/4/msaf077/8103706

Can be found at https://docs.rs/ska. We also have some tutorials available:

• From genomes to trees.
• From genomes to trees - Interactive tutorial in sandbox.bio.
• Filtering options.

Choose from:

1. Download a binary from the releases.
2. Use cargo install ska or cargo add ska.
3. Use conda install -c bioconda ska2 (note the two!).
4. Build from source

For 2) or 4) you must have the rust toolchain installed.

If you have an M1/M2 (arm64) Mac, we aren't currently automatically building binaries, so would recommend either option 2) or 4) for best performance.

If you get a message saying the binary isn't signed by Apple and can't be run, use the following command to bypass this:

xattr -d "com.apple.quarantine" ./ska

1. Clone the repository with git clone.
2. Run cargo install --path . or RUSTFLAGS="-C target-cpu=native" cargo install --path . to optimise for your machine.

This is a reimplementation of the SKA package in the rust language, by Romain Derelle, Johanna von Wachsmann, Simon Harris and John Lees. We are also grateful to have received user contributions from:

• Tommi Maklin
• Joel Hellewell
• Timothy Russell
• Nicholas Croucher
• Dan Lu

Optimisations include:

• Integer DNA encoding, optimised parsing from FASTA/FASTQ.
• Faster dictionaries.
• Full parallelisation of build phase.
• Smaller, standardised input/output files. Faster to save/load.
• Reduced memory footprint and increased speed with read filtering.

And other improvements:

• IUPAC uncertainty codes for multiple copy split k-mers.
• Uncertainty with self-reverse-complement split k-mers (palindromes).
• Fully dynamic files (merge, delete samples).
• Native VCF output for map.
• Support for known strand sequence (e.g. RNA viruses).
• Stream to STDOUT, or file with -o.
• Simpler command line combining ska fasta, ska fastq, ska alleles and ska merge into the new ska build.
• Option for single commands to run ska align or ska map.
• New coverage model for filtering FASTQ files with ska cov.
• Logging.
• CI testing.

All of which make ska.rust run faster and with smaller file size and memory footprint than the original.

• Sparse data structure which will reduce space and make parallelisation more efficient. Issue #47.

• Add support for ambiguity in VCF output (ska map). Issue #5.
• Non-serial loading of .skf files (for when they are very large). Issue #22.
• Alternative mixture models for read error correction. Issue #50.

• Use k > 63 (shouldn't be necessary? Let us know if you need this and why).
• ska annotate (use bedtools).
• ska compare, ska humanise, ska info or ska summary (replaced by ska nk --full-info).
• ska unique (you can parse ska nk --full-info if you want this functionality, but we didn't think it's used much).
• ska type (use PopPUNK instead of MLST 🙂)
• Ns are always skipped, and will not be found in any split k-mers.
• .skf files are not backwards compatible with version 1.
• Parallelisation is on a single node. If you want to parallelise across nodes for map see here. For build/align, you can use ska merge to combine .skf files at the end.