The PI plans to develop a new and robust hierarchical multi-resolution framework that can be used for assessement of model uncertainty, especially that associated with prediction of protein structure. The approaches also enable respresentation of the structure in ways that enable rapid searching of structure space. It is important to establish quality estimation methods for predicted structures so that they can be used wisely by knowing the limitations of the model. Protein tertiary structure prediction has made steady progress in the past decade. However, current prediction methods are still not capable of producing highly accurate models on a regular basis. Practical use of prediction methods by biologists is limited not only the accuracy of current prediction methods but also the lack of error estimation of the models they produce. Moderately accurate models are still useful for many purposes, including design of site-directed mutagenesis experiments and structure-based function prediction, if the possible error range is understood. Resulting quality assessment methods will also contribute to improvement of protein structure prediction methods. In addition, structure models of proteomes of model organisms will be constructed with quality assessment data and will be made available to the public through the Internet. The proposed project leverages Purdue University's efforts in interdisciplinary computational life science and engineering by training graduate students and undergraduate students of different backgrounds through interdisciplinary coursework and direct involvement in the project.
