Comparing metagenomes produces with different sequencing technologies

[dsher]

Hi all,

I would like to compare the abundance of my gene(s) of interest across many published metagenomes. Typically we would normalize the number of hits to the gene length and number of sequences (or mbp) to get relative abundances, and perhaps also normalize to the abundance of single-copy core genes to get an estimate of the gene number per genome equivalents. This has been done many times, but to the best of my knowledge all such comparison were between metagenomes sequenced using the same technology (e.g. within the GOS dataset, which is all Sanger).

My question is, can this approach be extended to compare across metagenomes produced with different sequencing technologies (e.g. compare the relative abundance in a marine metagenome sequenced with Sanger to human gut sample sequenced with Illumina)? Are such comparisons valid between samples with very different sequencing coverage? There must be some intelligent controls one would have to run... any suggestions welcome! Also, if anyone is aware of publications where such comparisons have been performed I would love to hear about them!

Thanks
Daniel

[Brian Bushnell]

My question is, can this approach be extended to compare across metagenomes produced with different sequencing technologies (e.g. compare the relative abundance in a marine metagenome sequenced with Sanger to human gut sample sequenced with Illumina)? Are such comparisons valid between samples with very different sequencing coverage?

I'm an author of an upcoming paper that does something along these lines. Generally, I think it's a bad idea to quantify across platforms like that. You will run into a lot of variable GC-related coverage bias, which is nonlinear and hard to compensate for.

[dsher]

Thanks Brian. Do these difficulties extend also to comparing between different versions of the same sequencing technique (e.g. between HiSeq and MySeq or between HiSeq or 454 versions as they evolve to provide different read lengths)?

Also, I would really be grateful if you could share the info on this publication (perhaps on this thread?) once it is publically available.

Cheers
Daniel

[Brian Bushnell]

Thanks Brian. Do these difficulties extend also to comparing between different versions of the same sequencing technique (e.g. between HiSeq and MySeq or between HiSeq or 454 versions as they evolve to provide different read lengths)?

To be honest, yes, unfortunately - different versions of Illumina chemistry have different GC biases; newer versions tend to have a lower bias than older ones. The read length shouldn't matter, but the chemistry does. I don't know if there is a substantial bias difference between chemistry versions on other platforms (454, Solid, PacBio), and probably some versions of Illumina chemistry are fairly equivalent to other versions and could be compared, but others definitely are not.

Also, I would really be grateful if you could share the info on this publication (perhaps on this thread?) once it is publically available.

It will be submitted this week, so hopefully that will be soon!