Domain Search Using Hmmer

Question

[Hari]

I used Hmmer search for finding domains from sequences,though HMMER is extremely fast and very useful,when i compared the results with the databases,HMMER identifies domains that are usually short of 1 or 2 residues,please help me how to overcome this.For example uniprot says domains starts from 'MKAM...' till '...ANDA';HMMER says its from 'AM....AN'.This happens because of the way HMM profile is defined or is it due to some parameter changes?

Thanks in advance

[Michael Kuhn]

Well, where do you get your HMM profile from? You can check the profile, see this previous question.

[Hari]

I got the hmm profile from PFAM.

[Hari]

I downloaded the sequences from pfam and got the hmmprofile

Answer 1

[Eric T.]

Are you using HMMer 3.0? The author blogged about the heuristic way HMMer 3 determines the sequence alignment:

http://selab.janelia.org/people/eddys/blog/?p=508

I've noticed it's common for the HMM alignment to treat the first or last 1-2 residues as inserts rather than matches, and it seems to be because of the kind of fuzzy probabilistic way the domain alignment is modeled. I'm not aware of any parameters you can change to affect the alignment envelope. Maybe this will be better in HMMer 3.1.

To make a slight tradeoff in sensitivity for consistency in your alignments, you could use RPS-BLAST instead. Try it out on a web server and download PSSM databases here:

http://www.ncbi.nlm.nih.gov/Structure/cdd/cdd.shtml

(It's also possible that the HMM profiles you used are not the same as Uniprot's.)

[Hari]

i used the HMMER 3 and it is really very good except for those missing residues.I will try also the cdd,but HMMER is better as I am looking for all (Rhogef) sequences in Uniprot with specific domains.

Answer 2

[Khader Shameer]

You may ensure the following aspects are consistent for the search you have performed vs. the one in the database you have tried.

• Version of HMMER (See: HMMER archive and recent version)
• HMMER parameters of programs like hmmsearch or hmmbuild (E-value, Cut-offs)
  
• Target HMM database that you used and the one in the web version

[Hari]

i used the recent version (HMMER 3.0) and the target database was uniprot and pfam was used for creating hmm profile