Automatic Literature-based Protein Annotation

描述(由申请人提供)： 对蛋白质功能的了解是生物医学研究的基石，生物医学研究是理解生物系统、疾病机制以及最终人类健康的基础。几十年的生物医学研究积累了大量以生物医学文献形式提供的此类知识。生物医学信息学的一项重要任务是从文献的自由文本中获取和表示知识，并将其转换为计算代理可以理解的语言，以便存储、检索和用于知识发现。目前，所有蛋白质注释都是手动分配的，不幸的是，这非常耗费人力，而且跟不上信息增长的步伐。事实上，随着几种模式生物基因组序列的完成，人工标注蛋白质已经成为生物医学文献中大量蛋白质和爆炸性信息量之间的一大瓶颈。在这一应用中，我们建议开发基于隐藏在生物医学文献中的功能信息来促进蛋白质功能的自动注释的方法。所提出的方法适应和扩展了概率语义分析、信息检索和机器学习方法的发展现状，这些方法是对自然语言文本中不确定性建模的原则性方法。该项目将为未来的自动注释系统开发算法构建块，以便当给定蛋白质的简要描述(例如，蛋白质名称和符号)时，它将能够检索有关该蛋白质的相关文献文章，从文章中提取生物学概念，并将该概念映射到受控词汇。我们预计，实现这些目标将带来影响更广的进展，不仅促进蛋白质的自动标注，还将促进生物医学文献索引--生物医学信息学的重要领域之一。高效的知识获取和管理将加强关于疾病机制和药物发现的生物医学研究。

DESCRIPTION (provided by applicant): Knowledge of protein function serves as a corner stone for biomedical research, which is fundamental for understanding biologic systems, the mechanism of disease and ultimately the human health. Decades of biomedical research has accumulated a great wealth of such knowledge available in the form of biomedical literatures. An important task of biomedical informatics is to acquire and represent the knowledge from free text of literatures and transform it to languages that are understandable by computational agents, so that the knowledge can be stored, retrieved and used for knowledge discovery. Currently, all protein annotations are assigned manually which, unfortunately, is extremely labor-intense and cannot keep up the pace of the growth of information. Indeed, with the completion of genome sequences of several model organisms, manual annotation of proteins has already become a major bottleneck between large number of proteins and exploding amount information in biomedical literatures. In this application, we propose to develop methods to facilitate automatic annotation of protein functions based on the functional information buried in the biomedical literature. The proposed methods adapt and extend the state of art probabilistic semantic analysis, information retrieval and machine learning methodologies, which serve as principled approaches to modeling uncertainties in natural language text. The project will develop algorithmic building blocks for a future automatic annotation system such that, when given a brief description of a protein (e.g., a protein name and symbol), it will be capable of retrieving relevant literature articles about the protein, extracting biological concepts from the articles and mapping the concept to a controlled vocabulary. We envision that achieving these goals will result in advances with broader impact which not only facilitate automatic protein annotation but also for biomedical literature indexing-one of the important area of biomedical informatics. The efficient knowledge acquisition and management will enhance biomedical research regarding the mechanisms of diseases and drug discovery.