You will recover a lot of ribosomal sequences despite using a method to decrease it. Normally you have >80% of total nucleic acids on cells that are rRNA, so even applying methods based on probe capturing or differential degradation among others, will leave a significant fraction of 28S sequences. One choice is to simply screen the presence of it using a two sequences blast comparison (one from the target organisms and another one, the dataset produced) and recover all the sequences producing a significant hit, and then assembly all of them to have a longer complete gene near 3800 bp. Or loading the single reads of the metatranscriptome to MG-RAST and using the SEED system, it will give you the list of sequences grouped as 28S, where it would be then possible to do an assembly to have a contig with the full gene.  

Hola Victor,

Another possibility is to use SortMeRNA (http://bioinfo.lifl.fr/RNA/sortmerna/) to pick out the 28S rRNA reads and then assemble them.

The precompiled version of SortMeRNA can slow if you have a lot of reads to process. The source distribution is easy to compile using a recent version of gcc with OpenMP to obtain better performance.

Hello Victor,

please keep in mind that assembling ribosomal rRNA gene sequences from metagenomic/metatranscriptomic data is especially problematic, because of the multiple large conserved rRNA-regions. Because these are conserved on DNA-level even between seperate taxa, the danger of assembling chimeras is very high. You may end up with sequences that actually consist of multiple species but may not always be recognized by chimera detection tools. If you do not have a paired end approach, you should not even bother to assemble the ribosomal genes. And even then I would only use relatively long reads (e.g. Miseq >250bp). Additionally, If you are assembling them with a k-mer based assembler, I would use relatively large kmers (>100bp) that span most conserved regions, to reduce the risk of chimeric assemblies. But even then I would regard the results with suspicion, if I cannot verify my sequences by another independent approach (e.g. sanger sequencing of 28S clones, or reference sequences which are known to occur in the sample), and would prefer to do that only for relatively simple (low diverse) communities.

Non-rRNA genes are less of a problem, because they are usually not that conserved on DNA level (even IF they are conserved on protein level)

Considering these complications, you may want to consider not assembling these rRNA reads after all. Instead you could use the above suggested tools (sortmeRNA is really very good) to just identify the 28S reads, then align them against a reference database to identify different hypervariable regions, and finally just focus on a specific hypervariable region in order to taxonomically classify these reads (basically mimmicking an amplicon approach for that data)