Pfam

What Pfam is

Pfam is a biological database that classifies proteins into families and domains, providing functional annotations and multiple sequence alignments. Established in 1995 by Erik Sonnhammer, Sean Eddy and Richard Durbin, it was designed to improve the speed and accuracy of semi-automated genome annotation, initially supporting work on the Caenorhabditis elegans genome. It is maintained by an international consortium and delivered through the InterPro platform at EMBL-EBI.

The resource classifies protein sequences into distinct entry types: families, domains, repeats and motifs. A domain acts as an autonomous structural or reusable sequence unit across different contexts, while repeats typically combine in tandem to form stable structures, and motifs present shorter sequence signals outside globular domains. Families sharing a common evolutionary origin, as supported by structural and sequence evidence, are further organised into broader groupings termed clans.

Pfam supports a broad spectrum of research across molecular biology. Experimental and computational biologists consult it to characterise unstudied proteins, structural biologists use it to identify targets for structure determination, and evolutionary biologists trace protein origins. It also catalogs domains of unknown function (DUFs), which represent conserved sequences across species whose biological roles remain to be determined.

How it is organised

Pfam entries are built using a two-tier alignment strategy. For each family, representative sequences drawn primarily from pfamseq (a non-redundant database of reference proteomes) and UniProtKB are aligned into a curated seed alignment. The HMMER software package converts this seed alignment into a profile hidden Markov model (profile HMM), capturing position-specific conservation across the family. This model is then searched against sequence databases, and matching sequences that exceed a manually defined gathering threshold are compiled into a full alignment.

Gathering thresholds are individually calibrated to maximise true positive detections while excluding false positives, which are monitored by checking overlaps against families that do not share a clan. Relationships between families within clans are guided by comparisons using tools such as SCOOP and structural information from the ECOD database. For searching queries, the InterPro web interface allows researchers to submit protein or nucleotide sequences, with DNA translated across all six reading frames prior to HMM matching.

Key facts

What you can use it for

Genome annotation

Annotate newly sequenced genomes by matching predicted protein or six-frame translated DNA sequences against profile HMMs to assign functional domains.

Remote homology detection

Detect distant evolutionary relationships in divergent sequences by weighting conserved positions through profile HMM searches rather than standard pairwise BLAST.

Structural biology target selection

Identify uncharacterised protein domains or boundaries to select suitable candidate targets for experimental structure determination.

Domain architecture analysis

Examine the arrangement, co-occurrence, and species distribution of functional domains, motifs, and repeats across protein sequences.

Strengths and limits

Questions