Computer Studies of Protein Structure and Function

Columbia University Health Sciences, New York NY

Investigators

Linked publications & trials

Abstract

DESCRIPTION (provided by applicant): The research described in the current proposal has the long-range goals of enabling the prediction of protein-protein interactions and the application of this knowledge to problems of biomedical relevance. An additional long-term goal is the fuller integration of Structural and Systems Biology. Specific Aims include: a) The development of three-dimensional structure-based methods to predict, on a genome-wide scale, whether and how two proteins interact. b) The integration of structural information with other sources of evidence as to protein-protein interactions. c) The application of the methods being developed to important biomedical problems. These research goals are motivated by a number of factors. First, cellular function is mediated by tens or even hundreds of thousands of protein-protein interactions yet these are generally hard to predict in advance or to measure accurately with high-throughput experimental techniques. A method that allows the computational prediction of such interactions would thus be of significant impact. Second, there are many more protein sequences than protein structures so it is necessary to find ways to amplify the information in protein databases if structure is to be fully integrated in genome scale research. The proposed research is intended to bridge this gap. A central element of the approach to be taken is the use of structural alignments to reveal novel functional relationships between proteins. Since structure is better conserved than sequence, these alignments can reveal new information. A novel structure-based method is introduced which exploits homology models and remote geometric relationships to amplify structural information. The evidence that is obtained is then combined with non-structural sources of evidence using Bayesian networks to yield a probability of whether two proteins interact. An important element in the proposed research strategy is the recognition that structural modeling on a large scale is necessarily imprecise so that it is necessary to use low resolution scoring functions for a given model that do not depend sensitively on atomic detail. Bayesian methods then allow the extraction of a clear signal from the underlying noise. The biological impact of the research will be enhanced through computational/experimental applications in areas where structural clues have the potential of discovering truly novel interactions. Validated predictions on adhesion proteins, nuclear hormone receptors and cytokine signaling proteins demonstrate the efficacy of the methodology. New applications to viral host interactions and to cancer signaling pathways are described.