improve EpiMutations() output

[jamesbaye]

Addressing issue #6

Added a cpg_ids column in tibble output.

If reduced_output = T, subset columns to ("sample", "chr", "start", "end", "cpg_ids").

I didn't include length or n_cpgs since they are redundant. Didn't include epi_id since this should probably be added when all samples have been processed.

[jamesbaye]

Latest commit addresses issue #7

[jamesbaye]

EpiMutations now produces a consistent tibble regardless of the choice of method.

When reduced_output=T, tibble columns are ("sample", "chr", "start", "end", "cpg_ids", "outlier_method", "outlier_score"). When reduced_output=F, extra columns provided by bumphunter are also carried over, including ("value", "area") which could be relevant for extra stats?

Updated docstring @return to precisely document the returned tibble object.

[yocra3]

Good work @jamesbaye! I think that the reduced output was a great idea. I have checked your code and I have some comments:

• Bumphunter output: I would not include either value or area in the output. valueand areaare measures of how consistent the DMR is and evaluate bumphunter results. In our implementation, we are using bumphunter to partition the genome and an outlier test to call the significance of the outlier. Thus, the statistics we are relying on the statistics that come from the outlier method and not by bumphunter. I do not think that value and area will help.
• Outlier test outputs: some methods might return two statistics, an estimate (outlier_score in your code) and a p-value. We should accommodate the table to hold both. As p-value will not be available for all, we could just set it to NULL or NA for the methods that do not have it. By memory, I would say that MANOVA and MLM are the two methods that will also report p-values.

[jamesbaye]

Thanks for the review @yocra3.

Re value/area: Are you suggesting that I remove the functionality of reduced=T? I thought it might be good that the user can check bumphunter statistics if they desired (reduced=F). Though the default (reduced=T) would not output them.

Outlier tests results: Okay, yes you're spot on. I'd just seen that yesterday. For now am I correct that each function still just reports one output? So the current reporting would work for now?

[carleshf]

@yocra3 , I think it makes sense to have the reduce_output argument to allow obtain the measurements form bumphunter since we will include methods that do not use it. Thereforem, for those that use it as partition algorithm it may be usefull to get its statistics (and for the methods that do not use it we may or may not report equivalent statistics, for instance, Berbosa .et al. will not report any statistics for the area partition process).

Therfore, I think:

• Mintain reduce_output (bu default set to FALSE) with the area and value (with some prefix bump_)
• Outlier, as you said

[yocra3]

Output: Leire is working on MANOVA and this function exports two parameters. Maybe, you can talk with her so you can integrate her function with yours.