SAKEIMA is a sampling-based algorithm for computing an approximation of the most frequent k-mers from a dataset of reads or a sequence. It's implementation is based on Jellyfish (version 2 https://github.com/gmarcais/Jellyfish ).
In order to replicate all experiments, the python script /scripts/download_ds.py automatically downloads and prepare the data we used in our experiments. Then, the python scripts /scripts/compute_exact_distances_parallel.py and /scripts/compute_sampling_distances_parallel.py can be executed to replicate our experiments. Before doing that, you need to compile SAKEIMA.
SAKEIMA has the same requirements of Jellyfish. You need autoconf, automake, libool, gettext, pkg-config and yaggo. Then compile using:
autoreconf -i
./configure
make -j 4
To execute SAKEIMA to count (canonical) k-mers using a random sample of a dataset of reads, you can use the python wrapper /scripts/run_SAKEIMA.py. The usage is the following:
usage: run_SAKEIMA.py [-h] [-k K] [-db DB] [-o OUTPUT] [-thr THR] [-dt DBTOT]
[-t THETA] [-l LAMBD] [-e EPSILON] [-ell ELL] [-d DELTA]
[-v VERBOSE]
optional arguments:
-h, --help show this help message and exit
-k K length of k-mers (>0)
-db DB path to input file (dataset of reads)
-o OUTPUT, --output OUTPUT
path to output file (counts of frequent k-mers)
-thr THR Number of threads to use for counting (>0, def. 1)
-dt DBTOT, --dbtot DBTOT
dataset size (>0). Computed if not given
-t THETA, --theta THETA
frequency threshold (in (0,1))
-l LAMBD, --lambd LAMBD
fraction of k-mers to sample (in (0,2))
-e EPSILON, --epsilon EPSILON
approximation accuracy parameter (in (0,1), def. theta - 2/dbtot)
-ell ELL size of bags to sample (>0, def. 1/theta - 1)
-d DELTA, --delta DELTA
approximation confidence parameter (in (0,1), def. 0.1)
-v VERBOSE, --verbose VERBOSE
increase output verbosity (def. false)
You need to specify the length -k of the k-mers and the path -db to the dataset (relative to the current folder); then, you need to provide in input the parameter theta using -t (--theta) or lambda -l (--lambda). The parameter theta is the frequency threshold for the frequent k-mers, while lambda is the fraction of k-mers to sample form the dataset. You can also fix both parameters. All the others parameters are optional; if not given, the algorithm automatically sets them. As example, you can provide the size (in terms of numer of positions) of k-mers in the dataset using the -dt option. If you do not, the algorithm automatically computes it.
To compute ecological distances between two sets of k-mers (and their counts) you can use:
jellyfish dump --dist mer_counts1.fa mer_counts2.fa
where mer_counts1.fa and mer_counts2.fa are two files containing the counts of the k-mers obtained from the dataset we want to analyze.
The documentation for computing the exact count of all the k-mers (using Jellyfish) is described in the /doc/ folder or at https://github.com/gmarcais/Jellyfish .