Bad assembly or bad sequence data?

[cyanoevo]

I'm trying to assemble a metagenome containing one cyanobacterial genome along with a whole bunch of commensal bacteria with the intention of separating out the cyanobacterial sequences. I have done an assembly using Abyss and then run the contigs through PhyloPythiaS. Very few contigs are coming out affiliated with cyanobacteria which is troublesome since that was the dominant organism in the culture. It looks as though the commensal bacteria (e.g. Pseudomonas) have assembled very well.

My question is, how likely is it that the assembler will generate chimeric sequences that obscure the organism of interest? Is there anyway of altering assembly parameters to compensate for this? Or is it more likely that my dna extractions have simply failed to work on the cyanobacteria?

[GenoMax]

So BBSplit analysis did not work that well?

[cyanoevo]

Not quite... It seemed to work, and mapped a portion of reads to the references, but when I assembled from the mapped reads there were still lots of contigs that were clearly from contaminants. I must say, some samples have worked better than others (I am working on several different strains) which is why I'm beginning to doubt my extractions...

[GenoMax]

So that is certainly one possible explanation. You could do another round of BBSplit with assembled contaminants to see if you can weed some additional sequences out. Then go back and do the assembly again.

[GenoMax]

Give SPAdes a try to see if it does a better job of assemblies. At least it may put the bacterial contaminants together better so you can remove them.

[cyanoevo]

Thanks, I'll have a go with SPADES. Currently, my workflow is something like this:

1) Trim adapters + poor quality reads - Trimmomatic

2) map reads to multiple cyanbacteria genomes - BBsplit

3) assemble mapped reads - ABYSS/SPADES

4) Separate contigs based on taxonomic affiliation - PhyloPythia

I was then thinking about using the initial reads to try and extend the cyanobacterial contigs. I've tried this using IMAGE (which didn't work) and PRICE (which ran out of memory on my 32 GB desktop). Currently the assemblies are generating huge numbers of contigs, many of which are short and obviously want to get this number down without throwing away useful data...

Does all this sound like a reasonable way of going about things? There is such a huge amount of available software out there it's hard to see the wood for the trees sometimes...

[GenoMax]

You may need to try and iterate between 2 and 3 to see if you can improve things. If the cynobacterial DNA is underrepresented in the current library then you may need to do another prep.

[Brian Bushnell]

When you say there are contigs from contaminants... what kind of contaminants are you talking about? The wrong strain, or the wrong phylum entirely, or synthetic lab molecules?