VectorizedKmers

VectorizedKmers.jl is a Julia package primarily designed for fast $K$-mer counting of biological sequences. The core idea is that $K$-mers with an alphabet size of $N$ are essentially integers in base $N$, and can be used as indices in a vector of size $N^K$ to count the corresponding $K$-mers.

This data structure can be used to quickly approximate distances between sequences. Notably, the squared Euclidean distance was used to approximate edit distance in this paper. The dot product has also proven to be a useful metric for comparing correlation between sequences.

julia> using VectorizedKmers, BioSequences

julia> kmer_array = count_kmers(dna"AACCGGTT", 2)
KmerArray{4, 2, Int64, Matrix{Int64}} with size (4, 4)

julia> kmer_array |> values
4×4 Matrix{Int64}:
 1  0  0  0
 1  1  0  0
 0  1  1  0
 0  0  1  1

julia> kmer_array[dna"AC"]
1

julia> kmer_array[dna"CA"]
0

The main downside of counting $K$-mers this way is that the arrays grow exponentially with respect to $K$. The 31-mer array of a DNA sequence would have a length of $4^{31} = 4,611,686,018,427,387,904$, which is equivalent to four exbibytes of memory, if the values are stored with 8-bit integers — which is just not feasible, really. Not only does allocating a lot of memory take up a lot of memory, but it can also take a substantial amount of time! This method of counting $K$-mers therefore works best for lower $K$-values.