Add inferelator

[AaronWatters]

This PR adds the Inferelator R code under a separate tree in the repository so we can build end-to-end comparisons between the refactor and the ChristophH R implementation.

The code is added separately so we can freely adjust it (for example removing reporting) without interference between code bases.

The PR also includes one test case which runs the Inferelator script using stripped down parameters. The test case only checks that the output files get created -- it does not validate the data inside the output file.

This base should allow us to modify the Inferlator R code to generate more easily analysed output formats in order to build better test cases and cross comparisons.

[codecov-io]

Current coverage is 97.80% (diff: 91.42%)

Merging #16 into master will decrease coverage by 0.14%

@@             master        #16   diff @@
==========================================
  Files            14         16     +2   
  Lines           880        912    +32   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits            862        892    +30   
- Misses           18         20     +2   
  Partials          0          0          

Powered by Codecov. Last update 8b0f7db...090ed5e

[nickdeveaux]

I'm just going to run this code locally on the model organisms and if the output is consistent with the R codebase I already have, I'll give a thumbs up

[nickdeveaux]

One thing: this R code base is not labeled as legacy anywhere. Would it be ok to move the top-level Inferelator directory into a folder called legacy, or rename it to Inferelator_legacy?

[nickdeveaux]

Hey Aaron, can we set this to true by default? This generates a condensed file of the network, with only the highest confidence edges. It shouldn't rely on any of the html, markdown, or knittr packages

[nickdeveaux]

I ran this locally and compared it to the output of the older R codebase. I see a lot of similarity between network summary files: you can compare summary_frac_tp_50_perm_1--frac_fp_100_perm_2_1.26.tsv in one run to summary_frac_tp_50_perm_1--frac_fp_250_perm_2_1.26.tsv and see that over 90% of the network edges are conserved (we can likely get a better similarity metric by also including the confidence score per edge).
However, looping over permutations of the priors is not a method that's being used by any current lab member, and there isn't a well-established method for integrating the output of this permutation loop, according to the people I've been chatting with (Mario, Dayanne, Kostya).

[dayanne-castro]

The goal of these option was to show that the inferelator is robust to
noise in the prior -- and that is exactly what you see...

On Thu, Aug 4, 2016 at 11:10 AM, Nick de Veaux [email protected]
wrote:

I ran this locally and compared it to the output of the older R codebase.
I see a lot of similarity between network summary files: you can compare
summary_frac_tp_50_perm_1--frac_fp_100_perm_2_1.26.tsv in one run to
summary_frac_tp_50_perm_1--frac_fp_250_perm_2_1.26.tsv and see that over
90% of the network edges are conserved (we can likely get a better
similarity metric by also including the confidence score per edge).
However, looping over permutations of the priors is not a method that's
being used by any current lab member, and there isn't a well-established
method for integrating the output of this permutation loop, according to
the people I've been chatting with (Mario, Dayanne, Kostya).

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#16 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AMZwNAtPCG3umjzW4ImJdZ596x36_Xd7ks5qcgDNgaJpZM4JSHcu
.