Ok. mg-Rast looks also like a very promising option.

I used Blast2GO, it's an ideal option for your objectives. However, it is an expensive option. If you are looking for a free trial try this:

https://server.t-bio.info/

Good luck!

Thank you Lo'ai Alanagreh . I will try with bioinfo platform from the link you kindly sent.

All best

Just in case someone has a similar question. This is what I did:

- Blastx the reads with Diamond agaist the RefSeq protein database, to get a set of reads which are probably coding regions of proteins.

- Translate the reads that matched in the previous step, using the same reading frame as in the matches in the first step, using Transeq.

- blastKoala the amino acid sequences.

It seems to have worked well.

I could never upload my reads to MG-RAST. The upload was constantly interrupted, and the files ended up being corrupt (wrong md5sum).

I think that you could also try to assembly the metagenome and then make the binning process, and use the bins within kofamKOALA.

Hi Mirna Vázquez-Rosas-Landa

Thank you for your reply. This is an interesting analysis route. I have seen many papers which first assemble the reads, and then do the annotation, but it feels like introducing unnecessary complexity to the analysis.

Intuitively, to annotate the unassembled reads directly sounds closer to a random sample from the whole community to me. I feel that by assembling first I could bias my results towards the highly abundant taxa (although their annotation would be more robust), by discarding the reads that could not be assambled. Also, some contigs (or parts of them) will have more coverage than others, and I would lose that information (or maybe use the coverage data to estimate their abundance?).

Also, I do not know how reliable the assembly and binning process is. I have done metaSpades assemblies, and maxBin2 binnings (after trying different methods). Many bins look like real genomes in CheckM, but others are less clear. If I discard the "bad" bins, I would be discarding the information from the (good) reads, due to limitation of the assembly and binnig, righ?

By annotating the reads directly, I would expect that the read counts mapped to a particular KO number would be related to the abundance of the genes in the original sample.

What do you think? Do you know of any comparisons between these two strategies (especially in diverse mock or real communities) and how they went?

Thank you again, and greetings!

Hi Eneas Aguirre-von-Wobeser

I think maybe it depends on the goals that you want to reach. If you map to the databases, I am not sure if you will identify something new.

However, I understand what you are saying about the possibility of bias your analysis; maybe that could potentially be solved by trying different binning software and try to recover the low abundant organisms.

I think you already went further with your analysis, and I think you are right; you will be discarding some good reads, but then maybe you can map those.

I don't know if someone has compared both strategies, probably they have; it would be good to see.