I understand the basic principles under which RPS-BLAST operates: search a protein (rpsblast) or nucleotide (rpstblastn) query against a database of protein profiles, a "reverse PSI-BLAST" search. My question has to do with the practical application of this tool to annotate conserved protein domains in bacterial assemblies.

Let's assume I have downloaded NCBI's Conserved Domain Database (CDD) and would like to annotate conserved protein domains in a bacterial genome I have just assembled from Illumina reads. I see two potential approaches.

  1. Run a gene prediction tool on the assembly, translate the predicted gene models, use rpsblast to search the translated sequences against the CDD profiles, and then transform the matches from local protein coordinates to global genome coordinates.
  2. Use rpstblastn to search the genome sequences against the profiles directly.

The latter option would be more compute intensive, since it requires computing a 6-frame translation of each genome sequence. But perhaps it is more sensitive in the presence of small misassemblies?

In practice, is either of these approaches preferred over the other? Are there any considerations I'm missing?

  • $\begingroup$ It feels like the arguments would be the same as doing blastx vs. gene prediction + blastp, and i agree with your assessment. perhaps another consideration is do you want to use this rpsblast (using PSSMs/PWMs) or something more complete like profile HMMs $\endgroup$ – Chris_Rands Jul 12 at 7:07

The first option. Using a pre-existing tool, such as PROKKA etc.

Diversity. Bacterial diversity is staggering and species such as E. coli and Salmonella, which are closely related diverged when mammals and dinosaurs/birds diverged, so searching nucleotide sequences is a bad idea as you will get nothing. Especially as you are hoping to find distant sequences that may be missed by the commonly used annotation tools.

Operons. Bacteria, especially thermophiles, often group their protein functionally in operons. Therefore, you can use the principle of guilt by association to guess what a mysterious protein does. It is extremely powerful. Say you have Voet & Voet handy, a pathway with a hole and a mysterious protein you can guess easily what it does.

Note, if you have a protein that you are particularly interested in, I would suggest making your own PSI PSSM file. The automated system cannot read the literature or do some quick checks to see when to stop, so may be overly conservative with the clusters.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.