Lachlan Coin 

Supervisor: Dr Richard Durbin

School: Wellcome Trust Sanger Institute

Protein domains are the structural, functional and evolutionary units of proteins. A protein can be regarded as a sequence of its domains. Given a new protein sequence, for instance from a genome project, the domains can be recognised on the basis of the similarity of sections of the amino acid sequence to known domain members. We have explored several methods for extending hidden Markov model (HMM) techniques for identifying protein domains.

The first method is based on language modelling from speech recognition. To increase word accuracy in speech recognition, language models are used to capture the information that certain word combinations are more likely than others, thus improving detection based on context. We demonstrate that a similar technique can significantly enhance protein domain recognition.

The second method is based on exploiting the fact that domains have uneven distribution across taxa. We use a taxonomic-specific association score to enhance protein domain recognition.

We have observed that closely related protein sequences which are each distantly homologous to a particular protein domain, are often either inconsistently identified as containing the protein domain, or not identified at all. Motivated by this observation, we have developed a phylogenetic method which scores a cluster of closely related proteins against a profile HMM. We show that this method can detect distant relationships not currently detectable by a standard profile HMM.