Metagenome analysis - first steps

[sfranzenburg]

Hi,
I have a few general questions how to analyze metagenomes.

Platform: Illuminia Highseq, 2x 150 bp PE reads with ~80 bp overlap
Aim: Get taxonomic representation and functional profile (e.g. KEGG)
Preferred tool: MEGAN 5

My questions to get started:

1. Should I first merge the paired reads by their overlap or should both be analyzed seperately? In MEGAN I could check a box for PE reads, so maybe not merging them serves a purpose. If I should merge, which tool do you recommend?

2. Should I collapse my sequences before blasting them. I could do that to group identical reads to OTUs. They get a new header than, stating how often that OTU was present. That would greatly reduce the blast time. Question is, can downstream tools like MEGAN deal with that?

3. Any recommended BLAST settings to get a good balance between accuracy and computational time?


Sorry for these stupid questions. I am not a bioinformatician. I am normally a try and error learner. But if every computational step takes several days, I better go the right way from the start...

Thanks,
Sören

[ctseto]

1. Combine the reads together, especially with that much overlap. Use FLASH, though I think Pandaseq does the same.

2. Depends on how much of your reads are unique. If a lot of your read content is not unique, it may not give you much.

I'm not sure what you mean by OTU's. OTU is Operational Taxonomic Unit and that comes up in the context of 16S data. fastx_collapser assigns a unique number to a unique read, but it's not an OTU.

3. An option you may want to think about is MG-RAST, or if you feel lucky, you can attempt a metagenomic de novo assembly, extract the open reading frames with prodigal and blast those.

[Brian Bushnell]

1. Should I first merge the paired reads by their overlap or should both be analyzed seperately? In MEGAN I could check a box for PE reads, so maybe not merging them serves a purpose. If I should merge, which tool do you recommend?

I won't comment on whether it is better to merge or not in your case, but we typically assemble the metagenome, then map to the assembly. The assembly gets annotated, and only reads that did not map to the assembly get merged, then annotated.

If you do merge, I suggest BBMerge; it's faster and more accurate than FLASH - orders of magnitude fewer false positives with similar merging rates, in my tests (which were done on 2x150bp libraries with average 30bp overlap).

Usage:

bbmerge.sh in1=read1.fq in2=read2.fq out=merged.fq outu=unmerged.fq hist=histogram.txt

[sfranzenburg]

Thanks to both of you, very helpful insight.

@Brian: Do you do de novo assembly with Trinity? How does it handle closely related bacterial species. My samples are gut tissues and I expect a variety of Acetobacter species in there.

[Brian Bushnell]

I have no experience with Trinity. We generally use Soap for metagenome assemblies. As for handling closely-related strains... well, the assembly will be worse in that case Still, you should end up with a lot of contigs much longer than reads. That means you have much less data to process, and the annotation/blasting/whatever will be more informative.

I've found normalization of reads helpful prior to metagenome assembly with Soap, especially when you have a lot of data - it yields a smaller assembly with less redundancy, better continuity, and more of the source reads map back to it. The link I gave also has a tool called BBNorm (bbnorm.sh) that can be used for normalization. But, that will tend to make it harder to resolve closely-related strains, so it depends on what you're trying to study.