HMMER is a free and commonly used software package for sequence analysis written by Sean Eddy. Its general usage is to identify homologous protein or nucleotide sequences, and to perform sequence alignments. It detects homology by comparing a profile-HMM (a Hidden Markov model constructed explicitly for a particular search) to either a single sequence or a database of sequences. Sequences that score significantly better to the profile-HMM compared to a null model are considered to be homologous to the sequences that were used to construct the profile-HMM. Profile-HMMs are constructed from a multiple sequence alignment in the HMMER package using the hmmbuild program. The profile-HMM implementation used in the HMMER software was based on the work of Krogh and colleagues. HMMER is a console utility ported to every major operating system, including different versions of Linux, Windows, and macOS.

HMMER is the core utility that protein family databases such as Pfam and InterPro are based upon. Some other bioinformatics tools such as UGENE also use HMMER.

HMMER3 also makes extensive use of vector instructions to increase computational speed. This work is based upon an earlier publication showing a significant acceleration of the Smith-Waterman algorithm for aligning two sequences.

A profile HMM is a variant of an HMM relating specifically to biological sequences. Profile HMMs turn a multiple sequence alignment into a position-specific scoring system, which can be used to align sequences and search databases for remotely homologous sequences. They capitalise on the fact that certain positions in a sequence alignment tend to have biases in which residues are most likely to occur, and are likely to differ in their probability of containing an insertion or a deletion. Capturing this information gives them a better ability to detect true homologs than traditional BLAST-based approaches, which penalise substitutions, insertions and deletions equally, regardless of where in an alignment they occur.

Profile HMMs center around a linear set of match (M) states, with one state corresponding to each consensus column in a sequence alignment. Each M state emits a single residue (amino acid or nucleotide). The probability of emitting a particular residue is determined largely by the frequency at which that residue has been observed in that column of the alignment, but also incorporates prior information on patterns of residues that tend to co-occur in the same columns of sequence alignments. This string of match states emitting amino acids at particular frequencies is analogous to position specific score matrices or weight matrices.

A profile HMM takes this modelling of sequence alignments further by modelling insertions and deletions, using I and D states, respectively. D states do not emit a residue, while I states do. Multiple I states can occur consecutively, corresponding to multiple residues between consensus columns in an alignment. M, I and D states are connected by state transition probabilities, which also vary by position in the sequence alignment, to reflect the different frequencies of insertions and deletions across sequence alignments.

The HMMER2 and HMMER3 releases used an architecture for building profile HMMs called the Plan 7 architecture, named after the seven states captured by the model. In addition to the three major states (M, I and D), six additional states capture non-homologous flanking sequence in the alignment. These 6 states collectively are important for controlling how sequences are aligned to the model, e.g. whether a sequence can have multiple consecutive hits to the same model (in the case of sequences with multiple instances of the same domain).

The HMMER package consists of a collection of programs for performing functions using profile hidden Markov models. The programs include: