fn6 0.1.3

Fast, efficient and scalable SNP distance calculation from disk.

FN6

Fast, efficient and scalable SNP distance calculation from disk.

FN5 reworked into Rust. Approximately 10x faster, easier to use and maintain, and adds checking for matching reference and mask in FN6 saves, all while retaining interoperability with FN5 saves.

Usage

All FASTA inputs can be optionally gzipped.

Usage: fn6 <COMMAND>

Commands:
  reference-compress  Reference compress a sample genome. This will create a .fn6 file that can be used for fast comparisons with other samples. The .fn6 file is a binary file that contains the compressed representation of the sample genome, as well as metadata about the reference and mask used for compression
  compute             Compute distances
  add-samples         Add some samples to existing samples. Only computes the extra distances required rather than all pairwise distances
  bulk-compress       Reference compress a set of genomes. Dumber than `ReferenceCompress` as it doesn't allow for setting specific IDs or output paths, but much faster as it can be parallelized across samples
  help                Print this message or the help of the given subcommand(s)

Options:
  -h, --help     Print help
  -V, --version  Print version

Reference compress

Reference compress a sample genome. This will create a .fn6 file that can be used for fast comparisons with other samples. The .fn6 file is a binary file that contains the compressed representation of the sample genome, as well as metadata about the reference and mask used for compression

Usage: fn6 reference-compress [OPTIONS] <REFERENCE> <MASK> <SAMPLE>

Arguments:
  <REFERENCE>  Path to the reference genome FASTA file
  <MASK>       Path to the mask file. The mask file is a text file containing the positions of the reference genome that should be masked (i.e., ignored) during the analysis. The positions are 0-based and should be separated by newlines
  <SAMPLE>     Path to the sample genome FASTA file

Options:
      --id <ID>          ID for this sample
      --output <OUTPUT>  Output path for the .fn6 file. If not provided, the .fn6 file will be saved in the same directory as the sample FASTA file with the same name but with a .fn6 extension
      --debug            Whether to print debug information to stderr
  -h, --help             Print help

Compute

Compute SNP distances

Usage: fn6 compute [OPTIONS]

Options:
  -r, --reference <REFERENCE>
          Path to the reference genome FASTA file. Only required if >=1 of the samples specified are fasta files
  -m, --mask <MASK>
          Path to the mask file. The mask file is a text file containing the positions of the reference genome that should be masked (i.e., ignored) during the analysis. The positions are 0-based and should be separated by newlines. Only required if >=1 of the samples specified are fasta files
  -s, --samples <SAMPLES>...
          Paths to sample files. Either .fn6, .fn5 or FASTA files. If FASTA files are provided, the `allow-fasta` flag must also be used. They will then be reference compressed on the fly (using the provided reference and mask) before distance computation
  -d, --directory <DIRECTORY>
          Directory to load from. Either .fn6, .fn5 or FASTA files. If FASTA files are provided, the `allow-fasta` flag must also be used. They will then be reference compressed on the fly (using the provided reference and mask) before distance computation
      --cutoff <CUTOFF>
          SNP threshold [default: 20]
      --fasta-extension <FASTA_EXTENSION>
          FASTA file extension to look for when loading from a directory. Only used if loading from a directory and if reference and mask are provided (i.e., if FASTA files need to be reference compressed on the fly). Default is "fasta" [default: fasta]
      --allow-fasta
          Whether to enable computation from FASTAs. It is recommended to pre-cache the reference compressed versions of the new samples to speed up computation
      --debug
          Whether to print debug information to stderr
  -h, --help
          Print help

Add samples

Add some samples to existing samples. Only computes the extra distances required rather than all pairwise distances

Usage: fn6 add-samples [OPTIONS]

Options:
  -r, --reference <REFERENCE>
          Path to the reference genome FASTA file. Only required if >=1 of the samples specified are fasta files
  -m, --mask <MASK>
          Path to the mask file. The mask file is a text file containing the positions of the reference genome that should be masked (i.e., ignored) during the analysis. The positions are 0-based and should be separated by newlines. Only required if >=1 of the samples specified are fasta files
  -s, --existing-samples <EXISTING_SAMPLES>...
          Paths to existing sample files. Either .fn6, .fn5 or FASTA files. If FASTA files are provided, the `allow-fasta` flag must also be used. They will then be reference compressed on the fly (using the provided reference and mask) before distance computation
  -d, --existing-directory <EXISTING_DIRECTORY>
          Directory to load existing saves from. Either .fn6, .fn5 or FASTA files. If FASTA files are provided, the `allow-fasta` flag must also be used. They will then be reference compressed on the fly (using the provided reference and mask) before distance computation
  -S, --new-samples <NEW_SAMPLES>...
          Paths to sample files to add. Either .fn6, .fn5 or FASTA files. If FASTA files are provided, the `allow-fasta` flag must also be used. They will then be reference compressed on the fly (using the provided reference and mask) before distance computation
  -D, --new-directory <NEW_DIRECTORY>
          Directory to load new saves from. Either .fn6, .fn5 or FASTA files. If FASTA files are provided, the `allow-fasta` flag must also be used. They will then be reference compressed on the fly (using the provided reference and mask) before distance computation
      --cutoff <CUTOFF>
          SNP threshold [default: 20]
  -f, --fasta-extension <FASTA_EXTENSION>
          FASTA file extension to look for when loading from a directory. Only used if loading from a directory and if reference and mask are provided (i.e., if FASTA files need to be reference compressed on the fly). Default is "fasta" [default: fasta]
      --allow-fasta
          Whether to enable computation from FASTAs. It is recommended to pre-cache the reference compressed versions of the new samples to speed up computation
      --debug
          Whether to print debug information to stderr
  -h, --help
          Print help

Bulk Compress

Reference compress a set of genomes. Dumber than `ReferenceCompress` as it doesn't allow for setting specific IDs or output paths, but much faster as it can be parallelized across samples

Usage: fn6 bulk-compress [OPTIONS] <REFERENCE> <MASK>

Arguments:
  <REFERENCE>  Path to the reference genome FASTA file
  <MASK>       Path to the mask file. The mask file is a text file containing the positions of the reference genome that should be masked (i.e., ignored) during the analysis. The positions are 0-based and should be separated by newlines

Options:
  -s, --samples <SAMPLES>...
          Paths to sample files. Either FASTA files or .fn6 files
  -d, --directory <DIRECTORY>
          Directory to load from
  -l, --list <LIST>
          Line separated file to read paths from
  -o, --output <OUTPUT>
          Output directory to write saves to. Useful when using `list` as it consolidates saves in a single directory. If not provided, the .fn6 files will be saved in the same directory as their corresponding FASTA files with the same name but with a .fn6 extension
      --fasta-extension <FASTA_EXTENSION>
          FASTA file extension to look for when loading from a directory [default: fasta]
      --debug
          Whether to print debug information to stderr
  -h, --help
          Print help

Performance

Both FN5 and FN6 produce the same results, but with condsiderable time differences. Below is a table of comparison between FN5 and FN6 performance on varying sets of Mycobacterium tuberculosis samples, randomly selected from the CRyPTIC dataset https://doi.org/10.5281/zenodo.16041005 All of the benchmarks were run on the same laptop with an Intel i9-13900H and 32GB RAM, directly from SSD.

N Samples (passing QC)	Comparisons	FN5 Reference compression	FN5 Compute pairwise matrix (per comparison)	FN5 Compute pairwise matrix (per comparison) no cutoff	FN6 Reference compression	FN6 Compute pairwise matrix (per comparison)	FN6 Compute pairwise matrix (per comparison) no cutoff	FN6 Compute pairwise matrix from FN5 saves (per comparison)	FN6 Compute pairwise matrix from FN5 saves (per comparison) no cutoff
100	4,950	1.394s	44ms (8.9µs)	175ms (35.6µs)	0.17s	7.93ms (1.60µs)	81.28ms (16.42µs)	9.29ms (1.88µs)	53.49ms (10.81µs)
250	31,125	1.698s	208ms (6.7µs)	1087ms (34.9µs)	0.33s	32.29ms (1.04µs)	280.36ms (9.01µs)	24.75ms (795.00ns)	270.21ms (8.68µs)
500	124,750	3.506s	628ms (5.0µs)	3905ms (31.3µs)	0.67s	84.04ms (673.00ns)	921.34ms (7.39µs)	48.75ms (390.00ns)	1010ms (8.13µs)
750	280,875	5.091s	1298ms (4.6µs)	8179ms (29.1µs)	0.92s	114.78ms (408.00ns)	2000ms (7.14µs)	91.94ms (327.00ns)	5580ms (19.85µs)
1000	499,500	6.735s	2287ms (4.6µs)	14299ms (28.6µs)	0.90s	152.88ms (306.00ns)	3480ms (6.98µs)	143.89ms (288.00ns)	8910ms (17.85µs)
1500 (1493)	1,113,785	11.363s	7481ms (6.7µs)	100481ms (90.2µs)	2.03s	300.10ms (269.00ns)	9130ms (8.20µs)	296.32ms (266.00ns)	22720ms (20.40µs)
2000 (1992)	1,983,044	16.377s	30805ms (15.5µs)	205160ms (103.5µs)	2.69s	487.53ms (245.00ns)	18390ms (9.27µs)	613.04ms (309.00ns)	42150ms (21.25µs)

All of the times to compute a pairwise matrix include the time taken to read the saves from disk. The time taken to load the saves from disk varies by tool, but is an important component of the real world runtime.

Mean time to load a save from disk across all above runs:

FN5	FN6	FN6 (from FN5 saves)
64.6µs	25.0µs	41.3µs

This also demonstrates the importance of utilising a SNP cutoff within the computation. Otherwise, per sample comparison times scale according to how disparate samples are.

Testing

For simplicity, testing is based on a dummy genome with 80 bases. This allows for known SNPs to be introduced and manually calculated

cargo test

Tree building

Optionally included in the Python package is code to produce a neighbour joining phylogenetic tree in newick format. This has a few requirements:

• A minimum of 4 samples within the comparison
• Distances are computed with no SNP cutoff

Example:

# Optional virtual environment
python -m venv env
source env/bin/activate

# Install fn6, both as a rust CLI tool and a python library
cargo install fn6
pip install "fn6[tree]"

# Create a table of distances with an arbitrarily high cutoff
fn6 compute -d path/to/some/samples -r path/to/a/reference -m path/to/a/mask --cutoff 9999999 > your-distances.txt

# Build a tree
find-neighbour-to-tree --input your-distances.txt --output your-tree.nwk