AIn0n / FMHD

Fast MinHash Distances algorithms collection

FMHD

Fast MinHash Distances algorithms collection

The project was described in the following B.Sc. thesis (in Polish): Efektywne algorytmy aproksymacji podobieństwa dla sekwencji genomowych.

Getting started

1. Build a project.
2. Add all filenames or directories with data to compare to a single file, e.g. data.txt.
3. Sketch e.g. ./FMHD sketch -l data.txt -o sketches.bin.
4. Compare e.g. ./FMHD ibf -l sketches.bin

build instructions

Just run build_FMHD.sh file. You will get ready to use the executable FMHD.

Configuration

Available flags:

• The -c argument determines how the algorithm processes data. For available commands see benchmarks, and the exact way they work is described in the thesis.
• The -l flag defines the name of the file that will be used as input for the algorithm. For the sketch command, this is a list of file names from which a sketch will be constructed. Other commands take pre-existing sets of sketches and return similarity results for each pair in the sketch. By default, results are returned as a triangle similarity matrix.
• Another option is -k, which allows you to specify the k-mer length into which the genome will be divided during sketching.
• The -o parameter is for the name of the output file storing the produced sketch
• Another parameter that can be configured is -t, indicating the number of CPU threads used for calculations.
• The -E flag allows to display of the data as an edge list, the default format is a triangular similarity matrix. Using this flag certainly slows down the algorithm's performance.

Used libraries

• CLI11
• fmt
• xxHash

Benchmarks

The following charts are a time comparison for pairwise sketch distance computation. Tests were applied for M = 1024. The results are the median from five independent computations.

Other programs mentioned in the benchmarks:

• Mash
• BinDash
• Dashing