Best platform for metagenomics?

[MikeT]

Hi all,
Our research group has lots of metagenomic-related projects ongoing: viral metagenomics, whole-community metagenomics and targeted metagenomics (mostly on 16S rDNA for prokaryotic diversity).
Since it seems that, in the future, Roche will cease production and maintenance of its 454, on which we have relied so far for our projects, we are looking for good sequencing platforms for our future projects. We might have access to an Ion Proton (or Torrent, not really sure about it), but we are also in contact with a good external service provider.
What would you suggest us to consider?
Thanks in advance

-MikeT

[krobison]

If you are mostly doing targeted 16S sequencing, then the PacBio RS II is generating interesting data in Circular Consensus mode. Clearly not everyone is going to buy one, but equally clearly you don't need to -- there are quite a few core labs which will run external samples & also some commercial vendors.

Barring that, the MiSeq is much more capable than any Ion platform, with 2x300 available now. Is the Proton supporting anything much longer than 100bp right now? I haven't been tracking this lately -- seems like the PGM may be shooting for 500.

[Brian Bushnell]

I would say that PacBio is extremely good for isolates - especially highly-repetitive ones! Our lab has been able to improve our fungal (which are highly-repetitive) assemblies dramatically by adding PacBio data; and create single-contig assemblies of microbial isolates, more accurately and cheaper than with Illumina-only data.

But for metagenomes, which may have a 1000000X difference in abundance between abundant and rare cells... HiSeq makes much more sense. If you only care about the few dominant organisms, any platform might suffice. But for a complete metagenome, you need the absolute most bases per dollar - which means not PacBio, and not MiSeq, but HiSeq. If you can afford multiple platforms it's worthwhile to sequence with both Illumina and PacBio and correct the PacBio reads with Illumina, but you still will only be able to assemble the rare organisms (say, 99% of the species, but 1% of the coverage) with Illumina data.

[AllSeq]

If you are mostly doing targeted 16S sequencing, then the PacBio RS II is generating interesting data in Circular Consensus mode. Clearly not everyone is going to buy one, but equally clearly you don't need to -- there are quite a few core labs which will run external samples & also some commercial vendors.

Barring that, the MiSeq is much more capable than any Ion platform, with 2x300 available now. Is the Proton supporting anything much longer than 100bp right now? I haven't been tracking this lately -- seems like the PGM may be shooting for 500.

Keith, PacBio can generate really high quality data, but does it generate enough reads for this application? How many reads do you shoot for in a 16S experiment?

[MikeT]

But for a complete metagenome, you need the absolute most bases per dollar - which means not PacBio, and not MiSeq, but HiSeq. If you can afford multiple platforms it's worthwhile to sequence with both Illumina and PacBio and correct the PacBio reads with Illumina, but you still will only be able to assemble the rare organisms (say, 99% of the species, but 1% of the coverage) with Illumina data.

Regarding assembly, here's another point I'd like to discuss, too: is assembly required to make sense of Illumina data?

-MikeT

[Brian Bushnell]

Not really. It depends on your goal. You can always map raw reads to functional or 16s databases to get an idea of what the community is composed of and what it's doing. With long enough reads, like 450bp insert 2x250bp libraries, you can annotate the overlapped reads to determine their (predicted) function. Ion Torrent reads may be long enough for this as well; I'm not sure. These methods need less depth than assembly.

[colindaven]

I'd suggest you have a look at public metagenomics data in places like the NCBI SRA. It's surprisingly easy to search through and find data from all types of sequencers currently employed. Then you can test the approaches directly.

I think HiSeq in general is pretty good for metagenomics, but long reads and long overlapping reads from the MiSeq look quite promising as well if you're doing gene prospecting and need long contigs.