Complicated Illumina data set - best way to assemble?

[jwag]

Hi all, I've recently acquired a large and fairly complicated genome dataset. No one in my department has done something of this magnatude, so I was hoping for some direction. The genome is eukaryotic and estimated to be approximately human-sized (~3 GB). We had it sequenced on an Illumina Hi-Seq2000 with the following:

Lane 1: Individual A, 300-1000bp insert, 100bp paired-end reads.
Lane 2: Individual B, 300-1000bp insert, 100bp paired-end reads. Also, two mate-pair libraries were sequenced of this individual, same read length as the others, one with a 5kb-8kb insert, and one with a 8kb-15kb insert.
Lane 3: 50/50 mix of individual A and individual B, 300-1000bp insert, 100bp paired-end reads.

Total was about 1.6 billion reads that passed the Illumina filter. The two individuals refer to two separate DNA preps from 2 separate organisms of the same species, and these individuals are likely diploid. The estimated coverage is ~50x.

So, as one would probably imagine, I'm having trouble finding the best way to assemble this genome de novo that 1.) Retains polymorphisms (ambiguities) at regions of heterozygosity and 2.) Allows for some degree of scaffolding at highly repetitive regions.

It's my understanding that I can't use Allpaths LG (which would retain those ambiguities) because I don't have a short fragment library. Also, we will likely be upgrading our server to at least 500 GB RAM soon, in order to handle this data.

Any thoughts on how to tackle this? Thanks in advance.

[GenoMax]

This is probably not the newest paper on this subject but should give you some options to consider: http://genome.cshlp.org/content/22/3/557

A complete list of relevant programs can always be found here: http://seqanswers.com/wiki/Software/list

[jwag]

I've looked at GAGE before. I suppose my broader question, is: is it even reasonable for me to try putting all 1.6 billion reads into an assembler and get a good output, especially considering polymorphism is probably present?

[GenoMax]

Since someone has already done the experiment you have to start somewhere.

I assume the 1.6 B reads are for all three lanes? You could start by trying to assemble the two individual genomes independently first. If there is a reference genome available for the specific (or a closely related) "organism" then you would at least have something to compare to. You could potentially use the reference genome for doing some alignments first to get an idea of coverage (or the potential lack of) in regions.

[jwag]

Thanks for the help-- yes the 1.6 B reads refers to all the lanes combined. I'll try assembling them separately first. Then perhaps I can map individual A to individual B to get an idea of polymophism.