Protein Family Databases for Automated Protein Domain Identification

Automatic identification and annotation of protein domains is a major challenge for genome sequencing projects. Simple transfer of the annotation from the overall most similar protein with a known function is relatively reliable for prokaryotic proteins, but often produces misleading and incomplete results for multi-domain proteins, which are common in higher organisms. An alternative approach is to classify protein domains based on matches to a precompiled database of protein domain families. A number of such databases are reviewed here, including an update on the Pfam database. The differences a user can expect to experience when using different databases for domain identification are illustrated by examples of known multi-domain proteins. The advantages and drawbacks of single-sequence versus multiple-alignment methods are also discussed. The degree of protein modularity was surveyed in the genomes of Caenorhabditis elegans, Saccharomyces cerevisiae, and Haemophilus influenzae by matching them to Pfam. While prokaryotic genomes typically have a small fraction of multi-domain proteins, that rarely contain more than three domains, at least 10% of higher eukaryotic proteins have multiple domains, many times with dozens of domains per protein chain.