Basic PGEN writer

[biona001]

This pr adds the write_pgen functionality which saves a dosage matrix into a PGEN file. Any suggestion is highly appreciated.

• In write_pgen, all hard-call genotypes are missing, while dosage values are saved in data track 4 of variant records
• Currently write_pgen is about ~6x slower than reading variant by variant, but I'm not sure how to optimize further. This line takes up roughly 70% of the time but I'm not sure if we can replace it with anything else
  

  PGENFiles.jl/src/write.jl

  Line 197 in ba32323

  storage[1] = hton(x)

• psam and pvar files are also outputted. I'm not sure if it's better to output a pvar.zst instead, but that would require adding CodecZlib.jl as another dependency

If this seems fine, I still need to add some documentation

[codecov-commenter]

Codecov Report

Merging #8 (ba32323) into main (931ca5d) will increase coverage by 4.15%.
The diff coverage is 88.42%.

@@            Coverage Diff             @@
##             main       #8      +/-   ##
==========================================
+ Coverage   68.51%   72.67%   +4.15%     
==========================================
  Files           9       10       +1     
  Lines         432      527      +95     
==========================================
+ Hits          296      383      +87     
- Misses        136      144       +8     

Impacted Files	Coverage Δ
src/PGENFiles.jl	28.57% <ø> (ø)
src/write.jl	88.42% <88.42%> (ø)
src/iterator.jl	56.66% <0.00%> (+3.33%)	⬆️
src/structs.jl	91.30% <0.00%> (+8.69%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[kose-y]

Thanks, this looks awesome. Are all the variant record types 0x40?

[biona001]

Yeah all variant record types are 0x40. Also apparently my tests are failing on windows, let me debug this...

[kose-y]

Have you checked if the output file works with Plink 2?

[biona001]

Sorry, haven't had much time to work on this pr this past week.

No I haven't yet.... To merge this pr, I still need to:

1. Check output works with PLINK2 and BASIL (a Lasso solver that I want to use)
2. Write some docs
3. Maybe improve efficiency. It just seems odd to me that writing is 6x slower than reading.

[kose-y]

The issue is that we are not closing the file we opened.

[biona001]

Sorry, do you mean the open(...) do io blocks? e.g.

PGENFiles.jl/src/write.jl

Line 54 in bb2e33a

open(pgen_filename * ".pgen", "w") do io

I think that syntax closes the io automatically when the block is done.

[kose-y]

Not necessarily. I think it’s `Pgen` struct not having a close function attached to it.

[kose-y]

@biona001 Please let me know if you think it is ready to be merged.

[biona001]

I have not checked if the result can be run/read by other PGEN readers. I haven't had time to do that yet

[kose-y]

Thanks for the update. By the way, adding close(p.io) (with p a Pgen struct) at the end of the tests might fix the test failures on Windows.