engleski

Hidden Markov models and function prediction for polyketide synthase domains

Polyketides (PK) and non-ribosomal peptides (NRP) are large families of biologically active compounds (e.g. the immunosupressants rapamycin and cyclosporin, the antibiotics erythromycin and penicillin). The complex enzymes that produce these substances - PK synthases and NRP synthetases - are divided into modules, which, in turn, consist of clearly defined functional units, called domains. Moreover, knowing the function of each domain enables one to determine the resulting compound. Genome sequencing projects of several PKS and NRPS hosts have revealed a large number of PKS and NRPS gene-clusters with unknown domain functions. It is of great interest to predict the function of these domains and scan the resulting compounds for biological activity or combinatorial biosynthesis potential. The prediction is obtained by considering a probabilistic model of the multiple alignment of domains in question, and combining it with a hidden Markov model (HMM)-based unsupervised learning method. This yields a division of the family of domains into several functional subfamilies. For example, we were able to predict two major functional groups for substrate recognition among acyltransferases, one of them showing further subdivision into two subfamilies. We can also recognise two major functional groups among ketoreductases with putative different stereochemistry during the ketoreductase reaction. Further applications of the method will be discussed.

Polyketides and non-ribosomal peptides; genome sequencing projects; hidden Markov model; HMM-based unsupervised learning; family of domains; functional subfamilies

nije evidentirano