de novo assembly of polyploid genomes?

[tsa]

Hi there,

looking for feedback on the performance of different assembly tools used for
polyploid genomes, based on 454 data, especially concerning assembly quality.

gsAssembler has a 'heterozygoteMode', but I'm unsure how well it works. Any
other tools you've used or can comment on?

[pengchy]

The same question. Any comments?

[tonybolger]

The same question. Any comments?

There's not a lot positive to say, unfortunately.

It is inherently difficult to assemble a complete homozygous genome using next gen data, diploid is harder still, and polyploid genomes even worse. AFAIK, such genomes are currently done using pre-separation of chromosomes in the lab.

Even in theory, the amount of data needed to disambiguate the various near-identical sequences into individual chromosome variants of a polyploid is massive.

First, you need enough reads to get a connected graph which includes all variants of the chromosome (and allows you to clean up errors vs variants). Then you enough long paired-end / mate pair data to find several unambiguous routes through the graph, with each route well supported by unique sets of spanning pairs of reads at every branching point. And this is all on top of all the ambiguity which comes from repeats and near repeats within each chromosome.

And i doubt that any tool out there today is really optimized for this purpose.

Net result - this is really hard to do.