Fast protein fold recognition and accurate sequence-structure alignment

We present two approaches to the sequence-structure alignment or threading problem: given an amino acid sequence and a protein structure, find the best mapping of sequence residues to structure positions with respect to some scoring system. Methods to solve this problem have two main applications: first, the recognition or identification of a plausible fold for a protein sequence of unknown structure out of a database of representative protein structures and, second, the computation of accurate alignments by improving on sequence alignments using structural information in order to find a better starting point for homology based modeling. We describe the application of these threading methods to a blind prediction of the structure of thymidine kinase (TK) of herpes simplex virus I: in combination with standard alignment and alignment evaluation methods implemented in our software package ToPLign, we were able to identify a model structure and to build a quite accurate partial model of essential parts of the structure including the active site.