martinjvickers / kast Goto Github PK

With m=1, d2s, d2star (and d2) all give the exact same as Martins python code.

With m=2, the numbers diverge. D2s is a smaller divergence, but with d2star we get a negative number which should never happen.

• Write recipe
• Create squashfs image
• Upload to singularityHub

Ideas?

• markov ordered limited to 3
• markov based kmer count to say 21 (e.g. (d2s and d2star))
• no limit on count based kmer counting (e.g. d2, manhattan etc.)

Create some known file output tests to use for complete unit tests.

• d2
• kmer
• manhattan
• chebyshev
• d2s
• d2star
• bray curtis distance
• normalised google distance
• dai
• hao

Maybe something like this to run the tests on various distros?

https://gist.github.com/jamesarosen/e29076bd81a099f0f72e

And for centos we could use docker on the ubuntu travis vm.

SAM/BAM embl, gbk etc.

Create some known file output tests to use for complete unit tests.

• d2
• kmer
• manhattan
• chebyshev
• d2s
• d2star
• bray curtis distance
• normalised google distance
• dai
• hao

• Create wrapper
• Publish into a galaxy toolshed

• dAI
• Hao

d2s-opt gave “nan” for everything.

The dai metric gives 0 for all tests, so something looks to be up with that one.

KAST gives a d2 score of -1.09891 when comparing a large (5000bp) fragment.

Use RAM to store pairwise comparisons. (TBH, why would you want to do a PW comparison on something THAT big, and if it is that big and you do it on disk, then it would take forever despite using no RAM).

The main thing to do here is ensure we check the amount of RAM FIRST and then fail fast with a warning.

The current test confirms that it runs without error, but doesn't check the result.

e.g. making it so that the correct options are available to user with the correct data. So now we've included protein data as an input (which I think is great) I do think we need to decide which options (e.g. reverse compliment, distance measures, kmer sizes etc) are available to the user. E.g. I could easily auto detect if we have nucleotide or protein data and then give a warning about the options for example.

This was vaugly mentioned in #25 which was focused on pairwise speed rather than RAM usage.

I don't know why this would happen generally, but KAST shouldn't seg fault for it.

I'm not sure of the best way to do this. Do it via the wiki as a tutorial and information, or should I force certain modes?

I think this is something to do with performing a count from within the markov_test function. Strangely this works fine for the old version.

I need to use Dna5String for protein and AminoAcid for amino acids

This could be in conjunction with function/unit tests;

• d2
• kmer
• manhattan
• chebyshev
• d2s
• D2S
• d2star
• D2Star
• bray curtis distance
• normalised google distance
• dai
• hao