The goal of the research described in this proposal is to develop a new generation of algorithms for comparative modeling, with a specific goal to improve the quality of models based on very distant homologies. This is a necessary step to provide a link between protein sequences coming from genome projects and understanding functions of these proteins, which can only be provided by the detailed knowledge of their three- dimensional structure. Enzymatic reactions, recognition of substrates, interactions between proteins-they all happen on the molecular level and whether we want to just understand or to modify, inhibit or enhance them, we need to look at and understand biological systems on the level of their molecular three dimensional structure. Unfortunately, none of the currently available algorithms is able to make the models more similar to the actual structures that they are to the templates they are built from. Actually in most cases the modeling processes partly destroys the similarity. The research described here aims to change it by combining two approaches: Development of empirical rules of how protein structures change in response to change in sequence and applying them to modify the template structure. Development of new tools for evaluation of three dimensional models of proteins In short, the plan is to use he first approach to generate a number of possible variants of the structure of the protein being modeled, while using the second approach to choose the best possible one. These overall goals will be accomplished by systematic analysis of changes in structures in families of homologous proteins to develop empirical rules of how protein structures changes in response to changes in sequence. The database of structures of homologous proteins at various levels of sequence divergence will be built and each structure will be decomposed into a hierarchy of subsystems built from smaller elements. This will allow seeking simple rules describing changes in structure, such as identification of """"""""pivoting moves."""""""" With a set of rules like that it will be possible to modify the structure of a modeling template to make it more similar to the final structure of the modeling target. Many possibilities will be generated and a system of model evaluation algorithms will choose the bet model among the as many as a thousand possibilities. A final goal of this proposal is to automate the algorithms to the point that they could be implemented in a fully automatic way on a WEB server. Improved algorithms, will be made publicly available on the group fold prediction server. Existing databases of sold predictions will be continuously updated and extended to include various interesting protein families.
Showing the most recent 10 out of 55 publications
