EXPERT HYBRID SYSTEM FOR PREDICTION OF PROTEIN STRUCTURE

基于知识的蛋白质二级和三级结构预测方法 结构仍然不够精确，不能保证将其应用于 实际问题 不同计算范式的整合 可能会大大提高其性能。 ORINCON 建议设计，构造和测试多专家混合系统 用于蛋白质结构的预测。 在这项工作的第一阶段，ORINCON将调整其MultEx范式 二次结构任务集成软件体系结构 预测. 拟议的系统将整合多项标准， 预测算法，并将另外包括非本地信息 通常不用于二级结构的蛋白质 预测，如氨基酸组成，部分实验 关于结构的知识等。这些信息与 氨基酸残基之间的长程相互作用， 被认为影响三级和二级结构。 的目的 第一阶段的工作是证明混合动力的优越性 方法和可行性，探索编码的经验模型 为了提供用于通过以下方式来提高预测准确性的手段， 包含在非本地数据中。 这项工作的结果将导致为第二阶段提出建议。 程序，其中蛋白质MultEx系统，一个商业上可行的 蛋白质二级结构预测软件包，将 将采用类似的方法， 三级结构预测方法的集成。 建议的商业应用 蛋白质结构预测的方法需要提取 从不断增长的蛋白质数据库中获取结构/功能信息 序列的 Protein MultEx系统将是一种商业上可行的 该产品应在新产品的开发和测试中找到应用 蛋白质结构预测方法，在结构生物学，并在 基于结构的药物设计

Knowledge-based methods for prediction of protein secondary and tertiary structure are still not accurate enough to warrant their application to practical problems. Integration of different computational paradigms may produce a substantial improvement in their performance. ORINCON proposes to design, construct, and test a multiple expert hybrid system for prediction of protein structure. In Phase I of this work, ORINCON will adapt its MultEx paradigm integration software architecture for the task of secondary structure prediction. The proposed system will integrate a number of standard prediction algorithms and will additionally include nonlocal information about the protein that is normally not used in secondary structure prediction, such as amino acid composition, partial experimental knowledge about the structure, etc. Such information is related to long-range interactions between the amino acid residues, which are thought to influence both tertiary and secondary structure. The aim of the Phase I work is to demonstrate the superiority of the hybrid approach and the feasibility of exploring the encoded empirical model to provide the means for improving prediction accuracy through the inclusion on nonlocal data. The results of this work will lead to recommendations for the Phase II program, in which the Protein MultEx system, a commercially viable software package for the prediction of protein secondary structure, will be fully developed and an analogous approach will be used for integration of tertiary structure prediction methods. PROPOSED COMMERCIAL APPLICATION Methods for protein structure prediction are needed to extract structural/functional information from the growing database of protein sequences. The Protein MultEx system will be a commercially viable product that should find application in development and testing of new protein structure prediction methods, in structural biology, and in structure-based drug design.