Multiscale Hierarchical Analysis of Protein Structure and Dynamics

Cornell University, Ithaca NY

Investigators

Abstract

The rapid growth in the number of experimentally determined protein structures has accentuated the need for theoretical and computational methods that can make this data useful. The proposed research aims to provide tools for important applications in computational molecular biology, such as protein engineering and structure-based drug design, as well as scientific explorations of structure-function relationships. This goal can be achieved through the ability to predict the conformational changes that occur when proteins and ligands interact. Protein engineering to alter enzyme specificity or to enhance stability under adverse conditions can also benefit from a predictive understanding of the motions involved in protein folding, unfolding, and catalysis. An attack on this wide range of challenging problems requires algorithmic breakthroughs, enormous computing power, and powerful techniques for visualization. Computational molecular biology uses information at the atomic level to study biologically relevant phenomena. Interpolating from the behavior of individual atoms to the collective behavior of complex biological molecules such as proteins presents a significant theoretical and computational challenge. Unfortunately, the available computational methods to study protein dynamics over long ranges of time are far from satisfactory due to a number of severe difficulties, including: (1) The time scales involved in theoretical/computational prediction span an enormous range: about fifteen orders of magnitude. (2) The conformation space of relevant protein shapes has very large dimension (> 10000) and the corresponding energy functions are quite jagged, making calculations expensive and difficult. (3) Algorithms for comparing/manipulating molecular shape are expensive in terms of both implementation difficulty and computational time. To resolve these difficulties the investigators are applying a variety of ideas centered on the theme of multiscaling: for time scales, for energy landscapes, and for measuring structural similarity at different levels of resolution. Algorithm development in the proposed area requires a unique blend of interdisciplinary expertise that is found within this team of investigators. The team includes a theoretical chemist who is an expert in protein dynamics, a theoretical physicist who is an expert in molecular biology and biophysics, and three computer scientists who bring expertise in geometric and combinatorial algorithms at both theoretical and experimental levels. This combined expertise is reinforced by a supportive interdisciplinary environment and excellent parallel computing resources. This research has three primary goals: (1) to broaden the range of time-scales for which meaningful Molecular Dynamics simulations of proteins are possible; (2) to improve our understanding of protein conformation space and the associated energy functions through the use of hierarchical multiscaling techniques; and (3) to enable the processing of proteins-as-shapes to proceed as easily as proteins are now processed as strings (over the 20-symbol alphabet of amino acids). Advances in these areas will significantly improve protein understanding, allowing computational experiments involving protein dynamics over wide ranges of time. Such improved computational abilities can potentially lead to important advances in the understanding of biology and the design of medicinal drugs.