Microbiome assembly and its normlization

[guptavipin142]

Hello,

I am working on 24 Human gut microbiome samples generated via illiumina hiseq, almost 800 gb data.
I have optimized and assembled the data using velvet but on different K mer, having assembly size; varying from 35 to 115mb. My query is there any need of justification to assemble all data on different k-mers. Is it ok or i should go for only one K-mer based assembly, also how do I proceed to assembly normalization.

[yzzhang]

For your assembly, you meant you pulled all data together and then assembled using velvet? I think this paper may be a good reference

[Zapages]

My suggestion is to try slew of different assemblers and different k-mer values.

In terms of genome assembly, the larger the k-mer the better the assembly as you are reducing redundancy in the sequence and increase coverage.

Now at the same time, it all depends on how much depth of coverage you want. If you have smaller k-mer value, then you will have more depth of coverage, but at the same time higher chance of mis-assembly of the reads.

So to over come this, I would suggest then taking all your assembled reads from the different k-mers and try to re-assemble them to form "super contigs".

This will help you achieve greater coverage (or reduce coverage to the specific genome by removing the same region that has been assembled multiple times) and reduce and bias that each assembler that might have.

Hope this helps a bit.

-Zapages

[guptavipin142]

Yes....
thanks for great help!!!!!!

[fanli]

IMO there are several newer assemblers out that typically outperform velvet in terms of both accuracy and speed. I typically use SPAdes for bacterial de novo assembly:

http://bioinf.spbau.ru/spades

see these as well for reference:

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3702249/

http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0107014

[lac302]

I've actually used Minia for metagenome assembly with good success.

Minia
http://minia.genouest.org/

Determining an optimal kmer size for a metagenome is tough. My suggestion would be to try several.

KmerGenie
http://arxiv.org/pdf/1304.5665.pdf

[guptavipin142]

IMO there are several newer assemblers out that typically outperform velvet in terms of both accuracy and speed. I typically use SPAdes for bacterial de novo assembly:

http://bioinf.spbau.ru/spades

see these as well for reference:

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3702249/

http://journals.plos.org/plosone/art...l.pone.0107014

Hi fanli,
All these seems to be genome assemblers and trained on genomics data I think.

[Brian Bushnell]

I have to say, Spades is slow even on a single microbe; I doubt you could run it on 800Gbp of metagenomic reads.

We had been using Soap and sometimes Ray for our metagenomes, but now are using Megahit which is faster and uses less memory than Soap.

also how do I proceed to assembly normalization

Can you clarify? I have written a normalization program to reduce high-depth reads prior to assembly, but I'm not sure that's what you are looking for.