Feature request: Alifold

[timoleistner]

Hi again,

Are there any plans to implement a wrapper for alifold? As I'm not familiar with C, I dont really know where to start but I would be thankful if you could lead me to the right direction. My current try looks like this

function alifold(sequences::Vector{AbstractString}, structure::AbstractString)
    alifold_output = LibRNA.vrna_alifold(sequences, structure)
    return alifold_output
end

Is the following needed somehow to

cstr_structure = Ptr{Cchar}(LibRNA.vrna_alloc(length(structure) + 1))
cstr_sequences = Ptr{Ptr{Cchar}}(LibRNA.vrna_alloc(sizeof(Ptr{Cchar})))

I'm almost certain this is not correct though. Do I also need to write the wrap_alifold function present in alifold.c? Do I need a specific struct to return the alifold information? Would be nice to have this implemented and I would like to help if I can.

Greetings, Timo

[marcom]

On v0.6.2 the following should now work:

fc = FoldCompound(["GGGAAAACCC", "GGCA-AAGUC"])
mfe(fc)
partfn(fc)
bpp(fc)

[marcom]

Closing for now, please reopen if there are any issues.

[timoleistner]

I've tested it and it seems like this line also returns true (or !false), if all strings in the msa are of equal length. https://github.com/marcom/ViennaRNA.jl/blob/ade41726aa07fda820d58666cab64f04845c11a2/src/ViennaRNA.jl#L63 This is rather interesting, and the reason is because the reduce function chains each evaluation after another. ((10 == 10) = true == 10) = false I have two suggestions for this line, and I've tested both:

using BenchmarkTools
test = [randstring(10) for _ in 1:100]

With the following results:

julia> @btime length(Set(length.(test))) != 1
  1.406 μs (11 allocations: 3.65 KiB)
false

julia> @btime !all(length.(test) .== length(first(test)))
  983.154 ns (6 allocations: 208 bytes)
false

Can you think of another prettier solution (they both feel kind of clunky)?

PS: Thanks alot for implementing it (and so fast too!)

[marcom]

Ups, you are right, that looks like a bug. I'll add a test for this and a fix

[marcom]

Thanks for testing this and finding the bug!

I implemented the check this way so it can exit early:

https://github.com/marcom/ViennaRNA.jl/blob/6a8fee28ac38285b6e29c187b1ca780565e3709c/src/ViennaRNA.jl#L63-L64

[timoleistner]

Another feature request, which I did not want to open another issue for, is to have the FoldCompound option for the sliding window size. Am I guessing correct that it does a global structure prediction with this default value (-1)? The options is called VRNA_MODEL_DEFAULT_WINDOW_SIZE. For longer sequences I think it would be cool to have this option for secondary structure prediction. Btw. do you have any idea how an appropriate value for this option would then be chosen?

[marcom]

As a quick possible alternative for folding longer sequences i would recommend to try LinearFold.jl. Maybe that covers your use case.

At the moment the structure prediction is always global. I agree that windowed predictions are still missing from ViennaRNA.jl (as are a few other algorithms).

I have opened a new issue #9 to discuss this further.

As for recommended window sizes, i think that depends on your use case, i.e. what kinds of structures do you want to be able to see?