There's a lot of slicing code in bioinformatics scripts I've seen written in python and R and so on, people extracting a gene or doing something else with a large sequence "just do it" and so it makes sense BioSequences goes to all the COW trouble for them to stop their scripts copying redundant data all over the place.
unsafe_setindex does, but it would be nice if we had a more generic mechanism, that uses the macros can do it for all mutating functions.
I've just been pushing forward on the new v2 api, which I might make a branch on BioJulia/BioSequences.jl soon.
The long and short of it is:
- The abstract
Sequencetype is now calledBioSequence{A}. - The type previously called
BioSequence{A}is nowLongSequence{A} Kmersare now a parametric type:Skipmer{U<:Unsigned, A<:Alphabet, M, N, K}. A Skipmer is a more generic type of Kmer, where in addition to K, you have two other numeric parameters M and N. They are more fully described here: https://www.biorxiv.org/content/early/2017/09/19/179960.full.pdf+html
The U parameter in the Skipmer type then allows you to pick the size of the UInt that stores the nucleotides.
There are aliases defined so you can use say
DNAKmer{27} in your scripts and such as shorthand and it will work.
I'm not sure if other people will use U, but I know in my group we like to use both UInt64 sizes kmers, and UInt128 sized kmers and compare results, as theres a tradeoff between number of kmers, and specificity/uniqueness, as you increase or decrease the length.
And we've found the 64bit limit quite critical in some cases.
Also, there's probably zero performance penalty with having the U parameter, so why not.
Your kmer iteration methods I think will remain, as for Kmers, I'm hard pressed to come up with something neater or faster, so we can dispatch to your iterators for Kmers where M and N == 1, and then the EachSkipmerIterator I've written for other generic skipmers.
Passing
Val{true} is described
I'll make that the next PR after the current one then!
I think it's just to show that with BioSequences you could store any element type, using an alphabet.
I think each char is stored in 8 bits, and so if you were doing bitops you could do things more quickly... what kind of things I've no idea
CharSequences are also mutable which String is not
But really it's something that is there, but isn't really used much right now.
I do envisage other alphabets in the future to store, for example a sequence of SNPs and that sort of thing as BioSequences would let you do that, but again there's no need or call for it right now.
I don't think the user cares that much about which internal checks BioSequences are doing. They probably just want a safe option (default) and an unsafe but fast option (where the documentation clearly states exactly which precautions they should take). However, I don't think it'sa good idea to make the COW-check parts of the bounds heck so it can be skipped with @inbounds, because the user might th
might reasonably assume that @inbounds only skill skipsa boundscheck and so might use it wrong
One of the best things about the COW-principle is that the user benefits without even having to know it's there.
@jakobnissen Yes I'm aware of that one, I think it's my fault, it's something in the bit-parallel site counting code I developed last year. I found a way to make some generic methods for bit-twiddling that work for all Unsigned sizes, with zero cost to performance, so I put them in Twiddle and also renamed some of them. I'll make a PR fixing this.
v2.0 of BioSequences, FASTX, Pseudoseq, and the newer packages (or newer versions of old packages) are being released through the "BioJulia package registry" now instead of the "General" registry.
