Hi guys,
May be I misunderstood something, but I think there could be a potential bug in the debruijn graph generation with super repetitive sequence. Imagine the following case:
AAAAATAAAATAAAATAAAATAAAATAAAATAAAATAAAATAAAA
It's a 45 length sequence with the following sequence A
+ AAAAT
* 8 + AAAA
.
I am curious what should be the right debruijn graph for this with 31 kmer size and strandedness as false (there is a potential problem with that too as msp and filter kmer assumes complementary definition of strandedness flag but assume we correct for it) i.e. I do want to canonicalize the kmers,
If I ran it correctly then this library gives the following gfa:
H VN:Z:debruijn-rs
S 0 AAAAATAAAATAAAATAAAATAAAATAAAAT
L 0 + 1 + 30M
S 1 AAAATAAAATAAAATAAAATAAAATAAAATAAAAT
P ENST00000375105.7|ENSG00000117215.14|OTTHUMG00000002701.1|OTTHUMT00000007683.1|PLA2G2D-001|PLA2G2D|2672|UTR5:1-59|CDS:60-497|UTR3:498-2672| 0+,1+,1+ *
NOTE: I added the P
flag and it was not available by default although L
and S
are available by default.
However twopaco, another library gives the following entries:
H VN:Z:1.0
S 15 ATTTTATTTTATTTTATTTTATTTTATTTTT
S 8 AAAATAAAATAAAATAAAATAAAATAAAATAAA
S 12 ATTTTATTTTATTTTATTTTATTTTATTTTAT
P ENST00000375105.7|ENSG00000117215.14|OTTHUMG00000002701.1|OTTHUMT00000007683.1|PLA2G2D-001|PLA2G2D|2672|UTR5:1-59|CDS:60-497|UTR3:498-2672| 15-,8+,12-,8+,12-,8+ *
If I follow the P values from the second gfa I can recreate the input fasta sequence. However, the size of the recreated sequence from the first gfa would be shortened by 4 bases, I think because of the repetitive sequences. Not sure how twopaco handles it (my hunch is by maintaining the reference sequence info ?) but I think this is a problem and mot sure how can we tackle it.
Let me know if you guys have any thoughts.