De-novo assembly of bacteria genomes - which tools?

[maasha]

Hi all,

I have been given the task of assembling three bacterial genomes.

Each have been sequenced with 454 with a read length around 250nt. But also with Solexa mate-pair (2.5kb space - 35nt reads) because the initial assembly of the 454 data gave too many contigs (using the Newbler assembler resulting in a couple of thousand contigs per genome - these being GC rich).

Which assembly software would be appropriate? And are there any recommendations on how to proceed with assembly having two types of data?

Cheers,

Martin

[BaCh]

Which assembly software would be appropriate? And are there any recommendations on how to proceed with assembly having two types of data?

You could try out release candidate 1 of MIRA 3 (or wait for rc2 which will be put online this weekend): http://chevreux.org/projects_mira.html (comes with extensive docs)

On top of my head: Velvet and Euler-SR will also work with 454 and Solexa. I certainly missed some. Have also a look at: http://seqanswers.com/forums/showthread.php?t=43

Regards,
B.

PS: I'm a bit biased as I wrote MIRA

[maasha]

Thanks BaCh,

So where to start? The Solexa Data or the 454? It occurs to me that the 454 - with the longer reads - are best for initial assembly - and then gaps can be closed with the mate-pair Solexa data.

And can you feed the resulting contigs of one of these assemblies into the software as reads?


Martin

[BaCh]

So where to start? The Solexa Data or the 454? It occurs to me that the 454 - with the longer reads - are best for initial assembly - and then gaps can be closed with the mate-pair Solexa data.

If you can't feed both read types at the same time, 454 will indeed give you a better initial assembly. Though things may change slowly with 76mers and soon the double.

And can you feed the resulting contigs of one of these assemblies into the software as reads?

Most of the time not: contigs are normally pretty long sequences where assemblers have problems with (they're currently built to expect reads up to 1-2k bases).

Try feeding the assembler with all reads at first. Compare what looks best to you. If that does not work at all, the resort to assembling Solexas, shred resulting contigs and feed those shreds alongside 454 reads to "normal" assemblers (Newbler, CABOG, MIRA, and others that grok 454).

At least that's what I'd do.

HOWEVER ...

First, I am skeptical that the Solexas will greatly reduce contigs in your project. First, having "thousands" of contigs for a small bacterium with 454 data is *extremely* unusual (unless your average coverage is a low single digit number). Something doesn't feel right.

Second, you wrote "GC" rich. Which might get problematic as Solexa has problem sometimes with GGC.G motifs. What I've often seen is this:

refseq .....GGCGGCGGCxxxxxxxxGCCGCCGGC.......

Now, the GGC motif in forward and reverse direction pretty much leads to a complete depletion of correct Solexa reads (they all have errors in the bases marked with x).

B.

[ewilbanks]

Try Velvet http://www.ebi.ac.uk/~zerbino/velvet/

has support for using both short (illumina) and long (454) reads in the assembly.

[maasha]

@BaCh,

Thanks,

First, I am skeptical that the Solexas will greatly reduce contigs in your project. First, having "thousands" of contigs for a small bacterium with 454 data is *extremely* unusual (unless your average coverage is a low single digit number). Something doesn't feel right.

Remember that the Solexa data is mate pair. So the question is also - is there any software that will take both single reads and mate pair reads as input?

@ewilbanks,

Thanks, velvet is on top of my list of tools to try (along with MIRA of cause).


Martin

[BaCh]

Remember that the Solexa data is mate pair. So the question is also - is there any software that will take both single reads and mate pair reads as input?

Any software able to treat paired-end should be very well able to handle single reads. It needs to anyway for cases where a mate is not present or unusable.

B.

[maasha]

Now I have battled extensively to assemble one bacterial genome sequenced with 454 and Solexa (mate pair).

454 data: 478840 reads covering 117689186nt with a mean length of 246 and 33% GC content (the other two genomes are the ones with high GC content).

Solexa data: Mate pair reads (distance ~2500nt) of 35nt and 44nt length in total 3308974 reads covering 130704473nt and GC content of 38% (can anyone explain this bias?).

Expected genome size: 2.8Mb

Using the Newbler assembler on the 454 data alone results in ~50 contigs.

Using Velvet on both data type I could at the best get down to 459 contigs using the below:

velveth test_both 31 -long M1_454.fna -shortPaired M1_solexa_paired.fna
velvetg test_both/ -ins_length 2500 -cov_cutoff 10 -exp_cov 41 -max_coverage 300

This is not too good (


(Installing MIRA is not going well because of the libboost requirement - can't libboost be included in MIRA?)



Martin

[BaCh]

Using the Newbler assembler on the 454 data alone results in ~50 contigs.
...
This is not too good (

I beg to differ: this is already not too bad for a 40x+ coverage. Now, having a 3MB data set that assembles into 700 contigs ... that's "not too good"

(Installing MIRA is not going well because of the libboost requirement - can't libboost be included in MIRA?)

Won't happen: including an almost standard library with a program is not a good idea, especially when all major distributions have it included. You could give the binary packages a try though, or don't they run on your machine?

Regards,
B.