Report metrics and phasing information

[egor-dolzhenko]

This pull request will:

• Change metrics file format to TSV
• Output optional phasing information into a single file (rather than creating a separate file for each repeat)
• Clean up Workflow.cpp
• Propagate DiplotypePaths to Diplotype name change

Example of a metrics file:

VariantId   Genotype    AlleleDepth
DMPK    13/15   28.38/27.84
ATXN3   20/20   20.42/22.72
HTT 14/17   21.00/16.35
HTT_CCG 12/9    17.89/20.74

Example of a file with phasing information:

LocusId Diplotype   Score
DMPK    (LF)(CAG){13}(RF)/(LF)(CAG){15}(RF) 700671
ATXN3   (LF)(GCT){20}(RF)/(LF)(GCT){20}(RF) 716594
HTT (LF)(CAG){14}(CAACAG)(CCG){12}(RF)/(LF)(CAG){17}(CAACAG)(CCG){9}(RF)    646181
HTT (LF)(CAG){14}(CAACAG)(CCG){9}(RF)/(LF)(CAG){17}(CAACAG)(CCG){12}(RF)    645285

Here (LF) and (RF) correspond to the left and the right flanks respectively; (CAG){14} corresponds to 14 copies of CAG; Score is the cumulative alignment score of all reads to the corresponding diplotype.

Notes

• The output of phasing information is turned off by setting the output file stream into the error state; data written to such streams are discarded. Are there any reasons to avoid this approach?

  std::ofstream file;
  file.setstate(std::ios::failbit);

• I am working on a draft of metrics/phasing documentation 😃

[egor-dolzhenko]

Minor code comment or 2. I have more thoughts at the design level:

1. Can the (LF) and (RF) elements in the phasing information be implicit? Are there cases where these wouldn't be part of the haplotype?
2. The concept of turning the output off by setting a bit on the ostream has a bad smell to it. A simple and transparent alternative is to have this controlled by bool -- what is the advantage of the ostream approach?

Thank you for the thorough comments!

• Insertions is one case where keeping (LF) and (RF) can be useful: Paths (LF)(INS)(RF) and (LF)(RF) correspond to presence and absence of an insertion (and "(LF)(RF)" seems better than an empty string). Since ExpansionHunter supports insertion calling, we might want to support them in REViewer too at some point. What do you think?
• Yes, agreed about the code smell. Originally I tried the ostream approach to avoid having ifs in multiple places. However, now that REViewer outputs a single file with phased haplotypes, perhaps phasing output should not be optional?

[ctsa]

Making it non-optional is even better. Maybe take this approach for now and add the option to turn it off once there's a good use case for this.

[egor-dolzhenko]

Making it non-optional is even better. Maybe take this approach for now and add the option to turn it off once there's a good use case for this.

Sounds good! I'll make this change and merge?

[ctsa]

Sure, or merge now, these are all just items to consider.