GitLab

Re-organization, equivalent code

Not using a pointer creates some problems with KmerSegmenter.
We should have a copy constructor in KmerSegmenter but it is a pain in the ***.

Not using pointers makes several copies of the same object. However the copies share the
same pointer to a KmerAffectAnalyser and that's the problem. When one of the copy is
deleted, its destructor is called which will delete the KmerAffectAnalyser and the other
copies won't have access to it, and it will also generate an error when the copies will try
to delete the KmerAffectAnalyser.

This has been solved by using pointers to KmerSegmenter everywhere.

As the 'operation' struct now contains the score, the S table is useless.

Now 'out_unsegmented' causes are output in segment.cpp, displaying affectation for each germline

This is an important change in how the heuristics works with multi-germlines.

Now the segmentation may run 2x to 3x slower (with -g -i), but we ensure that
the best germline is selected for each read. This improves the results,
in particular with mixed data from regular and incomplete rearrangements.

All the germline stats are now updated in core/windowExtractor.cpp.

The computer science department of the Université de Lille is is now CRIStAL

Useful for '-U' debug output.

Now KmerSegmenter.segmented_germline always contains the last tested germline, even
when it is not actually segmented.

core/segment.{h,cpp}: both V and J kmers must be above DETECT_THRESHOLD to flag the germline as 'detected'

This reverts the idea of 41b4f071, but still uses a 'detected' flag.
This flag tells us whether we should continue to the remaining germlines.
In the case of incomplete rearrangements, we certainly want to continue even if we have
many V or many J (but now both of them).
Tests d2a0dbd5 are now passing.

Note that we don't run this test when strand == 2, this should be changed.

Mathieu Giraud authored Nov 22, 2014 and Vidjil Team committed Nov 22, 2014

ABIGUOUS_THRESHOLD becomes DETECTED_THRESHOLD, such reads are flag 'detected'

This is cleaner: on one side we have the cause of unsegmentation, on the other we have the
flag 'detected' if there were enough V+J (now total V+J, we don't need to have both V and J).
Currently, the flag can be set when SEG_+/, TOO_FEW_V/J, or DELTA_MIN/MAX (but not STRAND).

We do not continue to the remaining germlines when 'detected' or STRAND.

This fixes a8d2c873. The idea was good, but not the code.

Getting UNSEG_AMBIGUOUS is not a normal situation. When it happens, we want
to exit and to report this sequence to the user, and not to segment by chance
the sequence on another germline.

When a sequence has both enough V and J affectations (above AMBIGUOUS_THRESHOLD),
it should normally be segmented. If not, the sequence is probably a chimera,
or the V and J repertoires are not well choosen. We now label this sequence
as AMBIGUOUS (instead of TOO_FEW_V or TOO_FEW_J).

As KmerAffectAnalyser is only used with KmerAffect, we do not need to keep the templates.
The code should be easier to read for C++ newbies such as the author of this commit.

Moreover, the argument 's' was not used

update json output for 2014.10.10 format

Again, this breaks some unit tests.

Now the MultiGermline contains only one germline.
The 'should' tests are still passing, but some unit tests are now broken.

Makes the compiler happy too.

This, together with commits d0ccbacd 4c5a51d8 and 61c5c497, solves all
compilation warnings except the ones coming from an unused parameter (which
can make sense).

Only results.before_left and results.after_right are signifiant to assert the segmentation

While CountKmerAffectAnalyser is more flexible, the heuristic implemented
in KmerAffectAnalyser is more efficient, that's why we use it in the KmerSegmenter