1998 Annual Report
Biological and Environmental Research

A Global Optimization Strategy for Predicting Protein Structure

Teresa Head-Gordon and Sylvia Crivelli, Lawrence Berkeley National Laboratory

Global optimization prediction from sequence (right) and crystal structure (left) of the -chain of uteroglobin. Methods like these can also be adapted to use soft constraint directives predicted by fold recognition algorithms to refine protein structures.

Research Objectives

We are developing a joint global optimization approach based on sampling, perturbation, smoothing, and biasing that has been quite successful working directly on the potential energy surfaces of small homopolymers, homopeptides, and recently, -helical proteins. Our overall strategy is to make good predictions of secondary structures by neural network techniques, and then manifest them as soft constraints to use within both a local optimization algorithm and as guidance within a global optimization framework.

Computational Approach

The use of soft constraints permits partial solution to the global optimization problem within a local optimization context by quickly refining -helices and -sheets when they are predicted with even moderate accuracy. The small-subspace global optimization techniques are concentrated on regions predicted to be coil, a category for which it is not possible to define a soft constraint, and should be particularly effective in resolving these regions.

Accomplishments

The developed strategy was parallelized and run on the T3E at NERSC using between 16 and 128 processors. A conservative estimate of the number of FLOPs needed to generate these results is

(102 N log N Flops/EF) x (104 N x M)EF

where N is the number of atoms, EF is the number of energy and force evaluations, and M is the size of the coil subspace, typically 2-10 degrees of freedom.

In the last year we have explored different parameterizations of the global optimization methods and tested their effectiveness on the prediction of a 70-amino-acid protein, uteroglobin (see figure). We have just begun a second all-helical target which is 104 amino acids, and we have several more helical protein targets to further test the robustness of our optimization approach. We have received follow-on funding from DOE to tackle the more difficult folding class of -sheet proteins.

Significance

The protein folding problem and the prediction of protein structure are the grand challenges in molecular biology. Understanding how and why proteins perform their evolved function is necessary both for reengineering defective proteins indicated in disease and for rational design of synthetic proteins relevant for biotechnical applications. The logical progression from amino acid sequence to protein structure to protein function makes timing critical for solving the protein structure prediction problem. As the Human Genome Project advances beyond mapping to sequencing the genome, we will be faced with an enormous database of amino acid sequences and a demand for protein structures for which x-ray diffraction and NMR methods will be inadequate.

Publications

S. Crivelli, T. Phillips, R. Byrd, E. Eskow, R. Schnabel, R. Yu, and T. Head-Gordon, "A global optimization strategy for predicting protein tertiary structure: -helical proteins," Proteins: Structure, Function, Genetics (submitted, 1998).


 INDEX  NEXT >>