HMMER is a very popular Profile Hidden Markov Model implementation.
For a detailed description of profile-HMMs, check out chapter 5 of our textbook. For a brief description, check out lecture slides.
From a practical point of view, profile-HMMs are good for two things :
- To search for members of protein families
- To search for distant homologs of single proteins
HMMER is also the basis of pfam.
HMMER is written in glorious C99 and has an exceptionally well written manual for both installation and usage.
HMMER is already installed on all SoIC and UITS computers such as silo, tank and Karst.
We only need to load it by doing module load hmmer.
Note that there is no excutable called hmmer but hmmer consists of multiple tools such as hmmbuild, hmmsearch, phmmer, and jackhmmer.
Alternatively, we could download the source code from the main website and compile, or directly get the binary version.
Finally, we could even use the EBI-hosted webserver to use hmmer without any installation.
Some of the tutorial files that are included with the HMMER download are included in this repository for your convenience.
In the case of a profile-HMM, our observation is not a sequence, but a multiple sequence alignment. Making the initial multiple alignment is not HMMER's job. We can obtain this alignment using various tools (clustal etc.) as HMMER is pretty forgiving with its input. The globins4.sto is an example multiple sequence alignment from the original hmmer tutorial. Let us examine this file.
hmmbuild builds a profile-HMM from a multiple sequence alignment.
Let us do hmmbuild globins4.hmm globins4.sto to accomplish just that.
The freshly built profile-HMM is now written in the file globins4.hmm.
Now that we have our model with, the basic operation is to search for it in a bunch of other sequences.
Note that these sequences are not aligned in any way and there could be quite a few of them.
For the sake of an example, we use 45 globin sequences provided in the HMMER tutorial.
hmmsearch is the tool and we can accomplish this task by simply doing hmmsearch globins4.hmm ../globins45.fa.
Let us examine the output.
Since this is a large output, we are probably better off redirecting it into a file.
Let us try doing hmmsearch globins4.hmm ../globins45.fa > globins4.out instead.
Let us examine the output using a pager such as less.
An alternative use of our new tool is to search for a conserved domain in a multi-domain protein.
The tutorial conveniently includes an example case.
fn3.sto is a multi sequence alignment for a finbronectin domain that is commonly found in various multi domain proteins.
We can easily build a model for this by doing hmmebuild fn3.hmm fn3.sto.
We could then search for this domain in a large multi-domain protein provided in the file 7LESS_DROME by doing hmmsearch fn3.hmm 7LESS DROME > fn3.out.
Let us examine what the output looks like this time around.
We can also leverage HMMER to sensitively search for distant homologs.
This is similar but reportedly superior to a BLAST search.
The executable for this operation is phmmer, which essentially builds a profile-HMM from a single protein sequence and search for it in a sequence database.
We can simply do phmmer HBB_HUMAN globins45.fa to search for human
jackhmmer extends on this feature by iteratively building a better profile-HMM from those sequences phmmer finds.
The usage would be identical to phmmer, such as jackhmmer HBB_HUMAN globins45.fa
Let us compare the outputs from phmmer and jackhmmer for the same query.
We can also build databases out of a bunch of profile-HMM models and query a single protein against that.
Now that we already have two models, we can combine them by doing cat globins4.hmm fn3.hmm > minidatabase and convert this into a searchable database by doing hmmpress minidatabase.
This will generate a bunch of binary files.
We can now query a protein against this database using hmmscan minidatabase 7LESS_DROME.
This is very useful if we have a large, possibly multidomain protein and we want to find out what kinds of domains it might have.
Finally, note that you can use hmmer similarly with DNA and RNA sequences.