POWRE: Combining Data Mining and Information Visualization Techniques with a Molecular Biology Sequence Similarity Database System

POWRE：将数据挖掘和信息可视化技术与分子生物学序列相似性数据库系统相结合

• 批准号: 9753283
• 负责人: Elizabeth Shoop
• 金额: $ 7.06万
• 依托单位: University of Minnesota-Twin Cities
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1998
• 资助国家: 美国
• 起止时间: 1998-01-01 至 1999-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9753283&HistoricalAwards=false
• 关键词: POWRE; Combining; Data; Mining; Information

The main objective of this project is to aid genome researchers with the task of elucidating patterns and clusters in large amounts of biological data. For genome researchers who are interested in comparing gene or protein sequences to the sequences within one genome or across genomes, this task involves executing hundreds of thousands of similarity searches that produce text output. This project involves the development of two specific software tools for visualizing and exploring the similarity data in a database of biological sequence similarity results. The first tool will be an Interactive Categorization Tool. This tool will display attributes of selected similarity database objects in a 2D scatterplot and enable dynamic manipulation of the display. This will enable the genome researcher to explore the attributes of similarities and categorize the similarities based on those attributes. For example, the genome researcher will be able to vary the input parameters of a function for computing the strength of each detected similarity and display a plot with the strength of each similarity shown as the color of each point, and the points situated in the 2D space based on score and statistical significance as the X and Y axes. The tool will enable genome researchers to dynamically manipulate the generation of higher- level concepts or categories for detected similarities (strong, marginal, and weak similarities as opposed to individual similarities with particular values of score and statistical significance that are more difficult to compare). This will lead to their ability to categorize hits as orthologous or paralogous, based on various attributes of the detected similarities. Score and p-value are not the only attributes that can be used -- the system is general enough that other attributes, such as percent identity, percent conserved, and length of alignment, among others, could be used in functions. Thus, genome researchers can cond uct exploration at different stages of the genome comparison research process. The second tool will be a Cluster Exploration Tool. Using the results from data mining techniques that cluster like sequences together, genome researchers will be able to visualize the similarities among the sequences in the clusters. For example, the tool can be used for a cluster of new unknown sequences that were found similar to members of a group of known sequences. The new sequences can be positioned as nodes on the left in a bipartite graph, and the known sequences that they are similar to can be positioned along the right. Lines drawn between the nodes, colored differently based on the strength of the hits, will enable the researcher to visualize the connectedness of the sequences in the cluster. Details about each sequence and each similarity in the cluster can be obtained from the DBMS. This will enable genome researchers to study groups of orthologous or parologous sequences. A key feature of these tools is that they will be 'thin' clients (often referred to as applets) that communicate with the underlying DBMS via queries formulated visually by the genome researchers. The use of Java- based components for these tools will enable them to be easily used and shared by the bioinformatics community and the genome research community. The development of these tools will demonstrate the feasibility of the thin-client approach that is the hallmark of the network computing architecture philosophy.

该项目的主要目标是帮助基因组研究人员阐明大量生物数据中的模式和聚类。 对于有兴趣将基因或蛋白质序列与一个基因组内或跨基因组的序列进行比较的基因组研究人员来说，这项任务涉及执行数十万次相似性搜索以产生文本输出。 该项目涉及开发两种特定的软件工具，用于可视化和探索生物序列相似性结果数据库中的相似性数据。 第一个工具是交互式分类工具。 该工具将在 2D 散点图中显示所选相似性数据库对象的属性，并启用显示的动态操作。这将使基因组研究人员能够探索相似性的属性，并根据这些属性对相似性进行分类。 例如，基因组研究人员将能够改变用于计算每个检测到的相似性强度的函数的输入参数，并显示一个图，其中每个相似性的强度显示为每个点的颜色，以及位于 2D 中的点基于分数和统计显着性的空间作为 X 和 Y 轴。 该工具将使基因组研究人员能够动态地操纵生成更高级别的概念或类别以检测相似性（强相似性、边际相似性和弱相似性，而不是具有更难以比较的特定得分值和统计显着性的个体相似性）。这将导致他们能够根据检测到的相似性的各种属性将命中分类为直系同源或旁系同源。得分和 p 值并不是唯一可以使用的属性 - 该系统足够通用，可以在函数中使用其他属性，例如同一性百分比、保守百分比和比对长度等。因此，基因组研究人员可以在基因组比较研究过程的不同阶段进行探索。 第二个工具是集群探索工具。利用将相似序列聚类在一起的数据挖掘技术的结果，基因组研究人员将能够可视化聚类中序列之间的相似性。例如，该工具可用于发现与一组已知序列的成员相似的新未知序列簇。新序列可以作为节点放置在二分图中的左侧，而与它们相似的已知序列可以沿着右侧放置。在节点之间绘制的线（根据命中的强度以不同的颜色着色）将使研究人员能够可视化簇中序列的连通性。有关簇中每个序列和每个相似性的详细信息可以从 DBMS 获得。这将使基因组研究人员能够研究直系同源或旁系同源序列组。 这些工具的一个关键特征是它们将是“瘦”客户端（通常称为小程序），通过基因组研究人员直观地制定的查询与底层 DBMS 进行通信。这些工具使用基于 Java 的组件将使生物信息学界和基因组研究界能够轻松使用和共享它们。 这些工具的开发将证明瘦客户端方法的可行性，这是网络计算架构哲学的标志。