Learning to Predict Protein Structure with a Representation Reduction Approach

学习用表征简化方法预测蛋白质结构

• 批准号: 9901617
• 负责人: Bruce Buchanan
• 金额: $ 6.5万
• 依托单位: University of Pittsburgh
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1999
• 资助国家: 美国
• 起止时间: 1999-05-01 至 2002-04-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9901617&HistoricalAwards=false
• 关键词: Learning; Predict; Protein; Structure; Representation

This standard award is for support of a PostDoc to work with the PI on his ongoing research whose objective is to apply advanced techniques from machine learning and information theory to study the problem of protein structure prediction. The techniques to be used include the utilization of intelligent ways of constructing useful composite attributes, and the dynamic reallocation of search effort as better approaches become recognized. Such techniques cannot be used blindly, but must be tailored to the special requirements of scientific discovery. The results of this work should have broad applicability to many aspects of automated and semi-automated scientific discovery, but the specific focus - understanding protein folding - is in itself extremely important to both protein chemists and society, because it will open new opportunities for research and medicine. Also, demonstrating a general technique for computer-aided scientific model design should help scientists in a variety of scientific disciplines, as they record more data than they can meaningfully analyze manually.

这个标准奖项是为了支持博士后与PI一起进行正在进行的研究，其目标是应用机器学习和信息论的先进技术来研究蛋白质结构预测问题。要使用的技术包括利用智能方法构造有用的组合属性，以及随着更好的方法得到认可而动态地重新分配搜索工作。这些技术不能盲目使用，而必须适应科学发现的特殊要求。这项工作的结果应该广泛适用于自动化和半自动化科学发现的许多方面，但具体的重点-理解蛋白质折叠-本身对蛋白质化学家和社会都极其重要，因为它将为研究和医学开辟新的机会。此外，展示一种计算机辅助科学模型设计的通用技术应该有助于各种科学学科的科学家，因为他们记录的数据比他们手工有意义的分析要多。