HMMER is used for searching sequence databases for homologs of protein sequences, and for making protein sequence alignments. It implements methods using probabilistic models called profile hidden Markov models (profile HMMs). Compared to BLAST, FASTA, and other sequence alignment and database search tools based on older scoring methodology, HMMER aims to be significantly more accurate and more able to detect remote homologs because of the strength of its underlying mathematical models. In the past, this strength came at significant computational expense, but in the new HMMER3 project, HMMER is now essentially as fast as BLAST. As part of this evolution in the HMMER software, we are committed to making the software available to as many scientists as possible. Earlier releases of HMMER were restricted to command line use. To make the software more accessible to the wide scientific community, we now provide servers that allow sequence searches to be performed interactively via the Web.

1. INTRODUCTION TO
HMMER
Biosequence Analysis
Using Profile
Hidden Markov Models
Anaxagoras Fotopoulos | 2014
Course: Algorithms in Molecular Biology

2. A brief History
Sean Eddy
 HMMER 1.8, the first public release of HMMER, came in April 1995
 “Far too much of HMMER was written in coffee shops, airport lounges, transoceanic flights, and
Graeme Mitchison’s kitchen”
 “If the world worked as I hoped, the combination of the book Biological Sequence Analysis and
the existence of HMMER2 as a widely-used proof of principle should have motivated the
widespread adoption of probabilistic modeling methods for sequence analysis.”
 “BLAST continued to be the most widely used search program. HMMs widely considered as a
mysterious and orthogonal black box.”
 “NCBI, seemed to be slow to adopt or even understand HMM methods. This nagged at me; the
revolution was unfinished!”
 “In 2006 we moved the lab and I decided that we should aim to replace BLAST with an entirely
new generation of software. The result is the HMMER3 project.”

3. Usage
 HMMER is used to search for homologs of protein or DNA sequences to sequence
databases or to single sequences by comparing a profile-HMM
 Able to make sequence alignments.
 Powerful when the query is an alignment of multiple instances of a sequence family.
 Automated construction and maintenance of large multiple alignment databases. Useful
to organize sequences into evolutionarily related families
 Automated annotation of the domain structure of proteins by searching in protein family
databases such as Pfam and InterPro

4. How it works
HMMER makes a
profile-HMM from a
multiple sequence
alignment
A query is created
that assigns a positionspecific scoring system
for substitutions,
insertions and
deletions.
HMMER3 uses Forward
scores rather than
Viterbi scores, which
improves sensitivity.
Forward scores are
better for detecting
distant homologs
Sequences that score
significantly better to
the profile-HMM
compared to a null
model are considered
to be homologous
Posterior probabilities
of alignment are
reported, enabling
assessments on a
residue-by-residue
basis.
HMMER3 also makes extensive use of parallel
distribution commands for increasing computational
speed based on a significant acceleration of
the Smith-Waterman algorithm for aligning two
sequences (Farrar M, 2007)

5. Index of Commands (1/4)
Build models and align sequences (DNA or protein)
hmmbuild
Build a profile HMM from an input multiple alignment.
hmmalign
Make a multiple alignment of many sequences to a common profile
HMM.

6. Index of Commands (2/4)
Search protein queries to protein databases
phmmer
Search a single protein sequence to a protein sequence database
Like
BLASTP
jackhmmer
Iteratively search a protein sequence to a protein sequence database
Like
PSIBLAST
hmmsearch
Search a protein profile HMM against a protein sequence database.
hmmscan
Search a protein sequence against a protein profile HMM database.
hmmpgmd
Search daemon used for hmmer.org website.

7. Index of Commands (3/4)
Search DNA queries to DNA databases
nhmmer
Search DNA queries against DNA database
nhmmscan
Search a DNA sequence against a DNA profile HMM
database
Like
BLASTN

8. Index of Commands (4/4)
alimask
Modify alignment file to mask column ranges.
hmmconvert
Convert profile formats to/from HMMER3 format.
hmmemit
Generate (sample) sequences from a profile HMM.
hmmfetch
Get a profile HMM by name or accession from an HMM database.
hmmpress
Format an HMM database into a binary format for hmmscan
hmmstat
Show summary statistics for each profile in an HMM database
Other Utilities

9. Basic
Examples
with HMMER
hmmbuild [options] <hmmfile out> <multiple sequence alignment file>
> hmmbuild globins4.hmm tutorial/globins4.sto
Most Used Options
-o <f> Direct the summary output to file <f>, rather
than to stdout.
-O <f> Resave annotated modified source
alignments to a file <f> in Stockholm format.
--amino Specify that all sequences in msafile are
proteins.
--dna Specify that all sequences in msafile are
DNAs.
--rna Specify that all sequences in msafile are RNAs.
--pnone Don’t use any priors. Probability
parameters will simply be the observed frequencies,
after relative sequence weighting.
--plaplace Use a Laplace +1 prior in place of the
default mixture Dirichlet prior.

10. Basic
Examples
with HMMER
hmmbuild [options] <hmmfile out> <multiple sequence alignment file>
> hmmbuild globins4.hmm tutorial/globins4.sto
Internal Use!

11. Basic
Examples
with HMMER
hmmsearch [options] <hmmfile> <seqdb>
Search a protein profile HMM
against a protein sequence
database.
> hmmsearch globins4.hmm uniprot sprot.fasta > globins4.out
Keynotes
hmmsearch accepts any FASTA file as target database
input. It also accepts EMBL/UniProt text format
-o <f> Direct the human-readable output to a file <f>
instead of the default stdout.
-A <f> Save a multiple alignment of all significant hits (those
satisfying inclusion thresholds) to the file <f>.
--tblout <f> Save a simple tabular (space-delimited) file
summarizing the per-target output, with one data line per
homologous target sequence found.
--domtblout <f> Save a simple tabular (space-delimited) file
summarizing the per-domain output, with one data line per
homologous domain detected in a query sequence for
each homologous model.
• The most important number here is
the sequence E-value
• The lower the E-value, the more
significant the hit
• if both E-values are significant (<< 1),
the sequence is likely to be
homologous to your query.
• if the full sequence E-value is
significant but the single best domain
E-value is not, the target sequence is
a multidomain remote homolog

12. Basic
Examples
with HMMER
•
•
•
•
phmmer [options] <seqfile> <seqdb>
search protein sequence(s)
against a protein sequence
database
> phmmer tutorial/HBB HUMAN uniprot sprot.fasta
jackhmmer [options] <seqfile> <seqdb>
Keynotes
phmmer works essentially just like
hmmsearch does, except you
provide a query sequence
instead of a query profile HMM.
The default score matrix is
BLOSUM62
Everything about the output is
essentially as previously
described for hmmsearch
jackhmmer is for searching a
single sequence query iteratively
against a sequence database,
(like PSI-BLAST)
Iterative protein searches
> jackhmmer tutorial/HBB HUMAN uniprot sprot.fasta
• The first round is identical to a phmmer search. All the
matches that pass the inclusion thresholds are put in a
multiple alignment.
• In the second (and subsequent) rounds, a profile is made
from these results, and the database is searched again
with the profile.
• Iterations continue either until no new sequences are
detected or the maximum number of iterations is
reached.

13. Basic
Examples
with HMMER
jackhmmer [options] <seqfile> <seqdb>
Iterative protein searches
> jackhmmer tutorial/HBB HUMAN uniprot sprot.fasta
• This is telling you that the new
alignment contains 936
sequences, your query plus 935
significant matches.
• For round two, it’s built a new
model from this alignment.
• After round 2, many more globin
sequences have been found
• After round five, the search ends
it reaches the default maximum
of five iterations

14. Basic
Examples
with HMMER
hmmalign [options] <hmmfile> <seqfile>
Creating multiple alignments
> hmmalign globins4.hmm tutorial/globins45.fasta
A file with 45
unaligned globin
sequences
Posterior Probability
Estimate

15. Smart(Hmm)er
Create a tiny database
> hmmpress minifam
> hmmscan minifam tutorial/7LESS DROME
> hmmsearch globins4.hmm uniprot sprot.fasta
> cat globins4.hmm | hmmsearch - uniprot sprot.fasta
> cat uniprot sprot.fasta | hmmsearch globins4.hmm -
Identical
> hmmfetch --index Pfam-A.hmm
> cat myqueries.list | hmmfetch -f Pfam.hmm - | hmmsearch - uniprot sprot.fasta
This takes a list of query profile names/accessions in myqueries.list, fetches them
one by one from Pfam, and does an hmmsearch with each of them against
UniProt

16. Latest Edition
Features
DNA sequence comparison. HMMER now includes tools that are specifically designed for DNA/DNA
comparison: nhmmer and nhmmscan. The most notable improvement over using HMMER3’s tools is the
ability to search long (e.g. chromosome length) target sequences.
More sequence input formats. HMMER now handles a wide variety of input sequence file formats, both
aligned (Stockholm, Aligned FASTA, Clustal, NCBI PSI-BLAST, PHYLIP, Selex, UCSC SAM A2M) and
unaligned (FASTA, EMBL, Genbank), usually with autodetection.
MSV stage of HMMER acceleration pipeline now even faster. Bjarne Knudsen, Chief Scientific Officer
of CLC bio in Denmark, contributed an important optimization of the MSV filter (the first stage in the
accelerated ”filter pipeline”) that increases overall HMMER3 speed by about two-fold. This speed
improvement has no impact on sensitivity.
Web implementation of hmmer

17. Available Online
phmmer
hmmscan
Hmmsearch
jackhammer
http://hmmer.janelia.org/search/hmmsearch

18. Advantages/Disadvantages
 The methods are consistent and
therefore highly automatable,
allowing us to make libraries of
hundreds of profile HMMs and
apply them on a very large scale
to whole genome analysis
 HMMER can be used as a search
tool for additional homologues
 One is that HMMs do not capture
any higher-order correlations. An
HMM assumes that the identity of a
particular position is independent
of the identity of all other positions.
 Profile HMMs are often not good
models of structural RNAs, for
instance, because an HMM cannot
describe base pairs.

19. More Information
http://hmmer.janelia.org
http://cryptogenomicon.org/

20. Thank you!
Algorithms in Molecular Biology
Information Technologies in Medicine and Biology
Technological Education
Institute of Athens
Department of Biomedical
Engineering
National & Kapodistrian
University of Athens
Department of Informatics
Biomedical Research
Foundation
Academy of Athens
20
Demokritos
National Center
for Scientific Research