Modeling and predicting all- [formula omitted] transmembrane proteins including helix–helix pairing

Abstract

Modeling and predicting the structure of proteins is one of the most important challenges of computational biology. Exact physical models are too complex to provide feasible prediction tools and other ab initio methods only use local and probabilistic information to fold a given sequence. We show in this paper that all- α transmembrane protein secondary and super-secondary structures can be modeled with a multi-tape S-attributed grammar. An efficient structure prediction algorithm using both local and global constraints is designed and evaluated. Comparison with existing methods shows that the prediction rates as well as the definition level are sensibly increased. Furthermore this approach can be generalized to more complex proteins.