Question on the de Bruijn Graphs

[aloliveira]

I ran a few times the Velvet, SOAP Abyss and found something intriguing (or that may be my ignorance) in all assembly. And here's my question:

1) Why two contigs which have approximately 68 base pairs of overlap are not merged together? I tried to think of some hypotheses linking the structure of the de Bruijn Graph. But i still not understand what happening here. I ran all assemblers with k= 51, and the database has millions of sequences with unique length of 54 bp.

Both contigs have overlapping kmers sufficient to generate a big merged contig and why this is not happening?

Thanks in advance,
André.

[seb567]

I ran a few times the Velvet, SOAP Abyss and found something intriguing (or that may be my ignorance) in all assembly. And here's my question:

1) Why two contigs which have approximately 68 base pairs of overlap are not merged together? I tried to think of some hypotheses linking the structure of the de Bruijn Graph. But i still not understand what happening here. I ran all assemblers with k= 51, and the database has millions of sequences with unique length of 54 bp.

Both contigs have overlapping kmers sufficient to generate a big merged contig and why this is not happening?

Thanks in advance,
André.

Hello,


The 68-base sequence is likely a repeat that is present at least twice in the real genome.

Therefore, joining the two contigs you described above will likely result in a chimeric contig -- something truly undesirable at any stage in a genome project.

It is usual for contigs to start and/or end with repeated sequences. Otherwise, they would be longer in the first place unless there was a true "data gap." (no sequences at all for a genomic region)


In my experience, ABySS is the best at avoiding misassemblies caused by repeats. So if ABySS says so, then you should not join your contigs.


Ray is also quite good (I work on Ray).



Sébastien Boisvert