Building Hidden Markov Model (HMM) for proteins

Question

[saadleeshehreen]

Hi,

I have 460 amino acid sequences for a specific protein family X. I want to make a HMM model for those sequences. I will use that HMM to search homologs of X / neighbour of X in some bacterial genomes. How can I make a HMM model ? Is there any software available? What is the step to do that?

Thanks in advance.

Answer 1

[toheitka]

The HMMER software (which is well documented) can be used to produce HMMs from alignments.

As for searching DNA with protein HMMs:

• nhmmer can be used to build nHMMs from nucleotide alignments. They can be used to query DNA.
• hmmer can be used to build HMMs from amino acid alignments. They can be used to query proteins.

The current HMMER manual (2015, version 3.1) states: "Still missing: Translated comparisons. We’d of course love to have the HMM equivalents of BLASTX, TBLASTN, and TBLASTX. They’ll come."

(Of course, you could always translate your genome in all frames , chop it up, and then screen it using protein HMMs. It is a bit ugly, you might run into some frameshift trouble, but maybe it works?)

[Jean-Karim Heriche]

To give a bit more details, to build a HMM for a set of proteins, the steps are:
- build a multiple sequence alignment with e.g. Clustal
- run hmmbuild (from the HMMER package) with the multiple sequence alignment as input

[saadleeshehreen]

Thanks. From their manual I came to know, I need to give .sto file as input to get a .hmm file. Does Clustal give .sto file as output? Or I have to use different software to convert my file after clustal?

Cheers

[Jean-Karim Heriche]

As far as I remember, hmmbuild can read alignments in several formats. Just check the docs.