Comparative modeling of mainly-beta proteins by profile wrapping

Abstract

The ability to predict structure from sequence is particularly important for toxins, virulence factors, allergens, cytokines, and other proteins of public heath importance. Many such functions are represented in the parallel [beta]-helix fold class. Structure prediction for this fold is a challenging computational problem because there exists very little sequence similarity (less than 15%) across the SCOP family. This thesis introduces BetaWrapPro, a program for comparative modeling of the parallel -helix fold. By estimating pairwise [beta]-strand interaction probabilities, a profile of the target sequence is aligned, or "wrapped," onto al abstract supersecondary structural template. This wrapping procedure may capture folding processes that have al initiation stage' followed by processive, interaction between the unfolded region and the already-formed substructure. This wrap is then placed on a known structure and side-chains are modeled to produce a three-dimensional structure prediction. We demonstrate that wrapping onto an abstract template produces accurate structure predictions for this fold (ill cross-validation: average C0 RMSD of 1.55 A in accurately wrapped regions, with 88% of the residues accurately aligned). In addition, BetaWrapPro outperforms other fold recognition methods, recognizing the .l-helix fold( with 1]00% sensitivity at 99.7% specificity in cross-validation on the PDB. One striking result has been the prediction of an unexpected parallel -helix structure for a. pollen allergen, and its recent confirmation through solution of its structure.