[Denovoassembler-devel] RE : Confused about coding -- 500k phiX simulation still fails

[Sébastien B. <seb...@ul...>]

Hello,

Can you rerun with -p tests/phix/phix_500k_1.fasta tests/phix/phix_500k_2.fasta

Otherwise I think your graph won't be connected enough because of the random number generator used to simulate reads.

Also, what is the read length ?


I added a system test for phiX and it works just fine on my branch master.

see system-tests/tests/phix

seb@fault:~/git-clones/ray/system-tests$ ./run-test-smp.sh phix|tee 1

I ran the assembly on my Samsung netbook with an Intel Atom processor.

[1,0]<stdout>:Number of contigs: 1
[1,0]<stdout>:Total length of contigs: 5384
[1,0]<stdout>:Number of contigs >= 500 nt: 1
[1,0]<stdout>:Total length of contigs >= 500 nt: 5384
[1,0]<stdout>:Number of scaffolds: 1
[1,0]<stdout>:Total length of scaffolds: 5384
[1,0]<stdout>:Number of scaffolds >= 500 nt: 1
[1,0]<stdout>:Total length of scaffolds >= 500: 5384
[1,0]<stdout>:
[1,0]<stdout>:Rank 0 wrote phix.Contigs.fasta
[1,0]<stdout>:Rank 0 wrote phix.Scaffolds.fasta
[1,0]<stdout>:Check for phix.*


                                                     Sébastien

> ________________________________________
> De : Eccles, David [dav...@mp...]
> Date d'envoi : 13 juillet 2011 18:40
> À : Sébastien Boisvert; den...@li...
> Objet : RE: Confused about coding -- 500k phiX simulation still fails
> 
> From: Sébastien Boisvert [mailto:seb...@ul...]
>> I think you generated too few reads.
>> That is why it fails.
>> You need to get the peak coverage at least above 20.
> 
> Okay, so I used the VirtualNextGenSequencer to create 500,000 reads, which
> has a substantially higher coverage:
> 
> http://pastebin.com/18SwshYE
> 
> It actually registers as having a peak now, which is nice, because it can
> then proceed with the assembly. The statistics are still a little odd,
> finding a peak at 1415 (neighbourhood counts 58,52,70,74*,56,72,68) when 2862
> has a higher count (70,54,72,82*,56,50,52), as does 2873
> (56,58,60,76*,64,76,64), but I'll put that down to smoothing -- it's close
> enough.
> 
> So now that I've muscled my way past the coverage, the assembly should be a
> breeze:
> 
> http://pastebin.com/xJJ9yycT
> 
> ... or maybe not. It [still] looks like it's failing on seed creation.
> 
> Would it make sense to include a coverage calculation in the seeding tests
> done by the workers (SeedWorker.cpp)? What happens if a vertex (with >5
> coverage) has a parent that has 1 edge in, one edge out, but the coverage of
> that parent is below 5?
> 
> -- David
>