Metagenomic Assembly from PacBio reads

[lmbrutscher]

Hi everyone,

I am trying to assemble a bacterial genome using PacBio reads and short illumina reads from a metagenomic pool of both eurkaryotic host and bacterial sequences.

In an ideal world, I would have been able to culture the bacteria and obtain pure DNA to sequence and assemble from, but its cultivation has proved challenging. So, instead we have a ton of metagenomic data sequenced obtained from whole host DNA. We have roughly ~70 million Illumina MiSeq paired end reads (2x150) and ~120,000 PacBio reads (5-20 KB). I suspect that I have roughly 3x coverage of the bacterial genome with the PacBio reads and ~20-30x coverage with the illumina reads (with 80-90% of the reads likely being host). This project is further complicated by the fact that the host genome is not fully sequenced, so I cannot remove known host sequence to simplify analysis and de novo assembly.

Could anyone recommend an appropriate set of programs/pipeline in order to isolate and assemble the bacterial genome? Thanks so much!

[rhall]

Hi,
There isn't one well developed approach to an experiment like this, I can think of a lot of things to try, but it is going to come down to a lot of manual assembly work.
My first observation would be that the pacbio coverage is a little low if you want to go for a hybrid assembly approach. With only 3x of data a lot of the genome will not be covered at all and the chances of the reads within that 3x spanning the important repeats, that will not be resolved in the illumina data, is slim.
I would concentrate on the illumina data, assemble with a standard metagenomic assembly algorithm, for a recent comparison http://journals.plos.org/plosone/art...l.pone.0169662. Then bin the microbial contigs by comparing with a database, using something like Kraken.
3x of pacbio data is going to be all but useless, but you could try manually aligning the pacbio data to the binned microbial contigs to see if you can get any useful scaffold information. The problem at 3x is that even if you do see a single read join two illumina contigs it is impossible to tell if that is a real connection or not. There is always a none zero chance of a biological chimera being formed during sample prep, a random event, so normally you would always require at least two reads to corroborate a connection.