Predictions of structure/function by PredictProtein

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): PredictProtein (PP) was the first Internet server for protein structure prediction when it went online in 1992 at EMBL. Ever since it has also been the most widely used structure prediction server. Since 1999, PredictProtein runs without financial public support at Columbia University. Limited CPU resources, prevent us from applying the best current methods; limited human resources prevent us from making the results more readily available to molecular biologists. PP differs from most other resources in two ways. Firstly, it tries merging a variety of tools into one single report. Secondly, a number of methods are unique to PP, e.g. the PHD and PROF methods for predictions of secondary structure, solvent accessibility, and transmembrane helices. Here, we propose a variety of technical and scientific solutions improving the functionality of PredictProtein. (1) The technical solutions address job and data handling, database update, user interface, web page layout, presentation of results, and directly linking original resources. (2) The systematic combination of methods requires evaluating these in parallel on identical tasks, e.g., at which level of probability should a signal peptide prediction override the membrane prediction. Our major focus will be on improving predictions for membrane helical proteins, developing methods predicting beta-membrane proteins, and on using structure predictions to more accurately infer functional information. Improving membrane predictions has become particularly urgent, since the recently solved high-resolution structures revealed that all existing methods were over-estimated. We hope that a combination of existing and new methods and a refinement of the respective alignments used will considerably improve prediction accuracy. To predict beta-membrane proteins, we want to explore a combination of novel prediction methods based on neural networks and similar systems with a Markovian-like model that implements the observed grammar in these proteins. As a particular example for using structural information to improve the reliability of inferring function, we propose to investigate the conservation of enzymatic activity.