GitLab

The 4 sequences added in 6503f6ae now give correct results.
All the 68 (now 72) test sequences are now passing on both strands.

This was a bug hidden since the first release of Vidjil.
It is now corrected, and the test added in 002adbd2 is now passing.

core/segment.cpp: Karlin-Altschul statistics, e-value computation with sequence sizes in the FineSegmenter

K and lambda are now roughly estimated.

There were several problems in the previous implementation. Now:
 - We handle separetly numbers of nucleotides in seq_{left,right} and in ref_{left,right};
 - There is no need to handle equality cases (if is the responsability of the segment_cost);
 - There is no need anymore to call dp.backtrack() (we directly use dp.best_score_on_i).
 - We also properly update del_D_left and del_D_right for VDJ recombinations

The tests added in 3c138657 are now passing.

no-modification

score_D may be not properly initialized, leading to spurious output
of D genes in the .vidjil file (as for TRB-VJ recombinations in bug20150909.fa)

The correct way to check for D is to use isDSegmented.

The test was more harmful than useful.
Corrects what was identified as bug20150909.fa.

This better reflects what we find in the reads:
UNSEG_TOO_FEW_V actually means that the J segment was detected, above
the e-value threshold.

We want to avoid confusion with the 's13' parameter in germlines.data

This allows us to further test if the sequence has been deleted and
avoids double free in case the FASTQ file is malformed.
Test added

Waiting for a true p-value computation, we just require here that
the score is at least the one of 10 matches.

This is now not a real p-value.

This could be used by the FineSegmenter as well.

No functional modifications.

The code does the exact opposite of what was told

"$(CXX) -M > dep.mk" also needs a "-std=c++11" flag

Instead of testing whether rep_4 is empty, we define
two different segmentation methods.

Selecting the segmentation method is now cleaner, and does
not require any string comparison. This allows to have (pseudo-)germlines with
the same locus but with different segmentation methods, as here
"xxx" for both MAX12 and MAX1U.

tests: tests .getIndexLoad() and .getProbability() with both AFFECT_UNKNOWN and AFFECT_NOT_UNKNOWN kmers