Recognizing protein folds with discriminative learning

描述(申请人提供)：计算生物学的基本问题之一是预测蛋白质的3D结构类别--即从其氨基酸的线性序列识别其折叠。拟议的项目旨在开发识别蛋白质折叠的计算方法和工具。第一个具体目标涉及建立并向科学界交付一个基于网络的、有区别的折叠识别软件引擎。该工具将首次以用户友好的形式实例化辨别折叠识别算法。在过去的5年里，这种类型的算法已经在科学文献中得到了描述和反复验证，但还没有简单易用的软件工具来将这种技术带给最终用户。第二个具体目标是通过利用问题固有的多类性质来改进现有的折叠识别算法。以前的方法分别对待每个折叠类别，因此牺牲了统计能力。这个项目将产生算法和软件，极大地提高我们从初级氨基酸序列识别蛋白质之间微妙的结构相似性的能力。

DESCRIPTION (provided by applicant): One (1) of the fundamental problems in computational biology is the prediction of a protein's 3D structural class -- that is, recognition of its fold from its linear sequence of amino acids. The proposed project aims to develop computational methods and tools for recognizing protein folds. The first specific aim involves building and delivering to the scientific community a web-based, discriminative fold-recognition software engine. This tool will instantiate for the first time in a user-friendly form a discriminative fold-recognition algorithm. This type of algorithm has been described and repeatedly validated in the scientific literature over the past 5 years, but no easy-to-use software tools yet exists to bring this technology to the end user. The second specific aim improves upon existing fold-recognition algorithms by exploiting the inherently multiclass nature of the problem. Previous approaches have treated each fold class independently, thereby sacrificing statistical power. This project will produce algorithms and software that dramatically improve our ability to recognize, from the primary amino acid sequence, subtle structural similarities among proteins.