OK, there are two possible concepts of functionality: one is for genes encoding enzymes acting in important metabolic steps, or another is about the genes that are actively transcribed vs the ones that are present and had been selected for a long time in the community. Is your analyses on total metagenomic/metatrasncriptomic datasets or is it on PCR amplicons targeting some genes in DNA and RNA (cDNA templates) environmental extracts or is one single organism? Depending of the complexity of the community (in soil is extremely diverse and a huge number of sequences will not cover the diversity of the metagenomes), then it is possible that you may run comparison of the reads that are found in both datasets. But this results will have always the limitation about how representative they are of the total RNA or DNA content. Is not an easy question to solve. As an starting point I would say, try to get the same amount of sequences on each dataset (metagenomic and metatranscriptomic to avoid normalizations and loosing information). And try to get as much as possible. As a guide, if you want to have a representative assembly of a genome, you need a minimum of 20X coverage with a 2x150 pair end library (read average 300 b). If an average bacterial genome is 4 Mbp, so you may need 80 Mpb of information for that single genome. Now, assuming that in soil community you have around 5000 different species, and 4 or 5 are above 1% in that community, to have a, let's say 20X representation of one single genome of one of the abundant species in the community, you may need a metagenomic dataset of 8000 Mbp (8Gbp, a full MiSeq flow cell). In the case of metatranscriptome is more difficult to do calculations as there the expression would be predominantly masked by rRNA over mRNA. And from the mRNA derived sequences (after RT PCR sequencing), can be classified with predicted function or mapped to the metagenomic dataset. 

Hello Ximena

Howard provides a good guide.  I have linked to a couple of metagenomic studies of phosphatase functional genes in grassland soil that we have performed here at Rothamsted. 

As standard practice we use paired-end sequencing, in the two linked papers we used 2x150bp and generated between 200 and 400 million reads per sample.  Generating this many reads is necessary because grassland soil microbial communities are both abundant and diverse.  OTU-based collector curves generated from these datastets trend as asymptotic, suggesting that most (though not all) of the diversity is covered.  We are now generating nearer to 700 million 2x250bp datasets with the latest Illumina sequencing technology, but have yet to analyse the data so cannot say yet whether the improvement is worth the extra effort (hardware, compute time etc.).

For metatranscriptomics - as Howard says - you will have to consider the rRNA issue.  We typically use subtractive hybridisation to remove as much of the rRNA as we can (the kits are pretty good) and are left with high quality mRNA.  If you take this approach you can generate fewer reads because the diversity is lower than DNA.  If you would like information on the subtractive hybridization then please let me know.

Article Phylogenetic distribution, biogeography and the effects of l...

Article Land-use influences phosphatase gene microdiversity in soils

https://www.reneshbedre.com/blog/sequencing-coverage.html