whole genome metagenomics - longer reads or more reads

[liux]

From data analysis point of view, if you had the option of 4 lanes of sequencing at 2x100 bp, or 3 lanes of sequencing at 2x150 bp, and you were trying to assemble a whole genome metagenomic sample, with about 20 major players, which option would you prefer? Longer reads, or more unique reads? and why?

[Brian Bushnell]

Longer reads are better for metagenomes, particularly if you go for overlapping (~270bp insert) reads; the ones that don't assemble can still be merged and potentially annotated. Longer reads also let you use longer kmers and yield a higher kmer depth per bp of sequence.

For a similar amount of data, of similar quality, longer reads are pretty much always better for assembly. Short reads are only better when doing high-sensitivity quantification, or examining really short things (e.g. microRNAs).

[luc]

Hi Brian,
Which assembler is making the best use of longer reads and of longer overlapping reads in your opinion?

Thanks

[Brian Bushnell]

I do a lot more assembly support than actual assembly, so I can't give you a conclusive answer there. But AllPaths-LG makes use of overlapping reads - theoretically, they don't need to quite overlap, as it will try to span a short gap with kmers - but right around the point of having a short overlap seems to be optimal. I don't personally use Ray, but one of my co-workers has been testing it extensively on metagenomes for a couple years, and feels that it makes excellent use of merged overlapping reads. From my testing, his Ray metagenome assemblies have much better contiguity and completeness metrics than other assemblers, such as Soap, though at the expense of more misassemblies and errors. Soap, on the other hand, gave inferior assemblies when fed merged reads compared to the raw data. But that's about all I can tell you. Maybe I can convince Michael to put up a merged read assembly comparison on his website, which is pretty nifty.

To be concise - I suggest trying Ray and AllPaths-LG if you have overlapping reads, but I have much more experience with Spades and Velvet.

[luc]

Thanks a lot Brian. I have not tried Ray in a while.
Somehow I have the impression that the current assemblers are not yet attempting to make full use of for example overlapping 300 bp reads (Allpaths LG has difficulties with longer reads).