Heterogeneous in situ data: Kernals, Distances and Trees

描述（由申请人提供）：该项目旨在为生物学家提供新的工具，帮助他们了解复杂系统，他们拥有不同来源的异质“原位”数据。这些数据呈现出许多层次的异质性，并且与需要整合到集成数据结构中的时空和先验信息同时出现。这次合作从收集过程的设计开始，并提供数据集成和分析工具，这些工具是围绕统计包R编写的，以及用JAVA编写的交互式图像分析程序GEMEDENT。该项目集中于两种特定类型的异构数据：由新的焦磷酸测序机提供的宏基因组数据和序列混合物，以及由自动显微镜提供的细胞图像数据。第一类异质数据是由斯坦福大学土木与环境工程系的Alfred Spormann收集的微生物土壤样本数据。该方案的重点是将贝叶斯计算应用于样本位置和采集序列数量的设计，然后使用谱多元方法将多样性指标作为表格（而不是摘要）进行分析，从而将数据结构纳入分解中。这些方法也将有助于研究来自HIV、细菌、病毒和癌细胞的混合数据。第二项研究的重点是免疫细胞与乳腺癌之间的相互作用，是与斯坦福大学血液学家彼得·李（Peter Lee）合作进行的。我们将分析染色淋巴结的显微镜图像数据。一个集成的图像分析系统能够自动检测许多不同细胞类型的位置和大小从染色图像。随机森林已被纳入图像分析系统，有效的交互式增强组件为用户提供了迭代学习过程的可能性，直到达到所需的精度水平。这些数据使我们能够推断肿瘤和免疫细胞之间的空间和动态相互作用。一名博士后将负责将同一患者的细胞数据与临床病史和微阵列表达数据相结合。异质性将通过基于谱分析、核方法和图形表示的探索性多变量技术来处理。

DESCRIPTION (provided by applicant): This project aims to provide biologists with new tools to help them understand complex systems for which they have different sources of heterogeneous 'in situ' data. These data present many levels of heterogeneity and come concurrently with spatio-temporal and prior information that need to be incorporated into integrated data structures. This collaboration starts with the design of the collection process and provides tools for data integration and analysis written around the statistics package R and an interactive image analysis program GEMEDENT written in JAVA. The project concentrates on two specific types of heterogeneous data: metagenomic data and sequence mixtures provided by the new pyrosequencing machines and cell image data provided by automated microscopes. The first type of heterogeneous data are microbial soil sample data collected by Alfred Spormann from Civil and Environmental Engineering at Stanford. The proposal focuses on applying Bayesian computations in the design of sample locations and number of sequences collected and then using spectral multivariate methods to analyze diversity indices as tables (instead of summaries), thus incorporating the data structure into the decompositions. These methods will also be useful in the study of mixture data from pyrosequencing HIV, bacteria, viruses and cancer cells. The second study focuses on the interaction between immune cells and breast cancer in a collaboration with Peter Lee, hematologist at Stanford. We will analyze data from microscope images of stained lymph nodes. An integrated image analysis system enables the automatic detection of the location and size of many different cell types from stained images. Random forests have been incorporated into the image analysis system and an effective interactive boosting component provides the user with the possibility to iterate the learning process until a desired level of accuracy is attained. These data enable us to infer the spatial and dynamic interaction between the tumors and the immune cells. A postdoctoral fellow will be in charge of combining the cell data with the clinical history and the micro-array expression data from the same patient. The heterogeneity will be dealt with by using exploratory multivariate techniques based on spectral analysis, kernel methods and graphical representations.