High-Accuracy Protein Models Derived from Lower Resolution Data

点击翻译按钮获取中文摘要

An outstanding international interdisciplinary team has been assembled that will bring a broad variety of expertise to bear on protein model building, bringing together researchers from chemistry, physics, computer science, mathematics, structural biology, and bioinformatics. The expertise ranges from quantum chemistry to machine learning, and from datamining to high performance computing. Input from collaborating NMR and crystallographers will be essential for validating the protein models. Improving abilities to model proteins can impact public health in important ways by enhancing our basic understanding of protein behavior and by facilitating a more efficient selection of protein targets for drug design. The overall goal is to improve a wide range of protein modeling approaches, both by developing new approaches, and by combining those previously been developed. The specific aims are to: 1) Improve existing comparative (homology) modeling and 2) Improve models obtained by fold-recognition and ab initio procedures to make them useful for molecular replacement. There will be some new methods development. Efforts are in four areas - databases, interaction potentials, conformational sampling, and optimization for combining approaches. We will develop ways to include constraints mined from sub-atomic resolution protein structures using a new HIRES Database (to include structures with resolution < 0.85 A). These will include a structure fragment database, as well as short-range distance distributions. These data can be used to compare modeled structures against the collected data. Uses of the high resolution data will ilclude selecting higher quality fragments to replace poor quality segments in the models, for mining interaction potentials, and as a source of a variety of other high quality information regarding protein structures. Better assessments of protein structural models will be developed, including the assessment of the quality of individual segments within a protein structure; the new metrics developed will be used for assessing the quality of computer-built models, crystal structures and NMR structures, and provide indicators of the expected quality of whole protein models as well as of its segments. New ways to sample protein motions will be pursued. Combining diverse methods will lead to significant gains in the computer modeling of protein structures. Extensive testing and validation will be carried out at each stage and in each part of the project to ensure large gains in model accuracy.