Protein Folding   <-   Research   <-   Sean Forman   <-   You Are Here

Abstract from Doctoral Thesis

A solution to the protein folding problem would predict the three dimensional structure of a protein from the sequence of its constituent amino acids. Current laboratory methods are inadequate when presented with the avalanche of amino acid sequence data newly available. This widening gap necessitates novel computational techniques. We will describe a hybrid approach combining a distributed search technique and a continuous optimization technique.

After a brief primer on protein structure and chemistry, we will introduce the various aspects of HOPS, an ab initio protein structure predictor. Following that, we will discuss the primary contributions of this thesis: selection of discrete torsion angles for combinatorial search of the protein fold space and the creation of non-local secondary structure.

The selection of discrete torsion angle sets (or clustering) requires a search through the space of observed protein torsion angles using a modified facilities location technique. This technique will generate any number of discrete torsion angle pairs. We give numerical results for solutions found using this method.

Non-local secondary structure involves the formation of parallel and anti-parallel beta-sheets from our discrete search space. The problem of beta-strand alignment is formulated as a non-linear equality constrained least-squares problem (tweaking). The constraint is replaced with a linear approximation and an iterative technique involving Lagrangian multipliers is introduced to solve this problem.