Mapping vs denovo assembly for non-SNP genomic variations

[jpjp]

Hi Guys,

even though I have been working with NGS for a while it is not clear to me why mapping onto a reference genome is preferred to denovo assembly to decipher genomic variations like CNV, inversions, translocations etc.

I guess one issue is the difficulty in confidently assembling the entire sequence from short reads. But given that most of the new data is paired end and the read lengths are increasing steadily, what other problems do you see with denovo assembly? If this issue is resolved, then would denovo assembly identify the genomic variations more accurately and confidently?

Thoughts and insights are welcome, especially in the light of BGI's push in this direction.

http://www.bio-itworld.com/2011/07/26/bgi-researchers-publish-structural-variation-pipeline-push-for-de-novo-assembly.html

[jpjp]

Just bouncing it.

Nobody has an opinion on it??

[Shlomo Blum]

Nothing, eh?

I'm working with bacterial genomes, and although I have plenty of refs out there (E. coli...), I still wonder what's best, de novo or mapping, since we aim to search for novel genes.

Mapping to K-12 is giving us about 20% of unmapped reads (the novel genes?). Not bad. But still, would de novo give us more?

SB

[krobison]

Mapping is much faster & requires a much smaller compute power. You also have the annotations already mapped to the reference.

How close is your organism to the reference genome? At a minimum, I would assemble the unmapped reads to see what falls out.

Even with highly variable genomes, you can expect major failures from mapping only. As a trivial example, imagine mapping an F+ lambda+ E.coli K12 strain onto MG1655 -- you'd have a lot of interesting biology falling out as unmapped reads.

[Shlomo Blum]

They are the same species, but not the same phylogeny. Anyway, even if mapping to a reference, the idea is to assemble de novo whatever remains unmapped.

The point is, how much a difference does it make to assemble either way? I was told that mapping improves coverage depth, but you may lose information. I just don't get if you lose information because of unmapped reads are "lost" or because mapped reads would actually be assembled differently in a de novo context.

Also, going for a reference mapping option, how should I know ("bioinformatically", as much as possible) where in the genome (chromosome X plasmid) fall the de-novo-assembled-unmapped reads?