Once you have identified a domain of interest with Pfam, you can use the corresponding hidden Markov model (HMM) to search against a sequence database with hmmsearch. A potentially much faster solution would be to use PoSSuMsearch2.

As for searching contigs or open reading frames, I'd use both, so to minimize the possibility that you miss something. If a domain is coded by several exons, you might miss it when searching the contigs (I am assuming that you get transcripts from your gene identification software). You might also miss domains when searching the gene models, because these models might just be wrong, or exons or whole genes might be missing. I don't know the extent of this danger, but better safe than sorry.

Once you have identified a domain of interest with Pfam, you can use the corresponding hidden Markov model (HMM) to search against a sequence database with hmmsearch. A potentially much faster solution would be to use PoSSuMsearch2.

As for searching contigs or open reading frames, I'd use both, so to minimize the possibility that you miss something. If a domain is coded by several exons, you might miss it when searching the contigs (I am assuming that you get transcripts from your gene identification software). You might also miss domains when searching the gene models, because these models might just be wrong, or exons or whole genes might be missing. I don't know the extent of this danger, but better safe than sorry.

As far as i understand, you have already identified the domains and now you need to search for them in your metagenomic data.

There are 2 ways to predict known domains (given that the HMMs exist), and these both ways include the InterProScan search. InterPro is a database of protein domain databases and searching it gives normally reliable results (because you search simultaneously several DBs).

Way 1 is to download ItperProScan, but it is hard to do for seveal reasons.

Way 2 is to use Blast2GO (google it), which automatically sends each protein sequence to InterPro and writes down the result. It takes some time, but is very comfortable to use. I use it for genome annotations. So all you need to do is to make a multifasta of your predicted protein sequences and start the Blast2GO (select immediately the InterPro option, you can skip all the other steps like GO annotation, etc).

I agree with Marc that it is more reliable to scan also the genomic sequences not to miss a wrongly annotated gene. But whether to make this step depends on how sure you are in the gene prediction. If you have doubts - try another predictive tool. To my knwledge, the protein domains are normally searched on the amino acid level (and the HMMs are made for amino acids).

Thank you Mark and Ekaterina, I'll take a look at these options!

Hi, I suppose hmmsearch would do the job for you. Also you can give it a try with CLC genomic workbench :)

Good luck