Capturing protein sequence-structure specificity using computational sequence design.

摘要: It is well known that protein fold recognition can be greatly improved if models for the underlying evolution history of folds are taken into account. The improvement, however, exists only such evolutionary information available. To circumvent this limitation families have a small number representatives in current sequence databases, we follow an alternate approach which benefits including recreated by using sequences generated computational design algorithms. We explore strategy on large database templates with 1747 members from different families. An automated method used to these templates. use backbones experimental structures as fixed templates, thread self-consistent mean field approach, and score fitness corresponding semi-empirical physical potential. Sequences designed one template translated hidden Markov model-based profile. describe implementation method, optimization its parameters, performance. When native were tested against library profiles, class, fold, family memberships majority (>90%) correctly recognized E-value threshold 1. In contrast, when homologous same library, much smaller fraction (35%) recognized; structural classification sequences, (with accuracy >88%). Proteins 2013; © 2013 Wiley Periodicals, Inc.