Compressing and Analyzing Microarray Images for Genetic Information Extraction

生物信息学是信息技术在研究和分析生物和基因数据方面的应用。功能基因组学是人类基因组计划的第二阶段，它将功能赋予DNA序列，并基于新的微阵列技术迅速发展。微阵列图像使数千个基因的同时表达成为可能，并掌握着理解基因调控和相互作用、癌症等疾病的遗传途径以及寻找对某些药物最敏感的亚群的关键。它们广泛应用于学术界和工业界的实验室。微阵列成像技术的出现使图像处理在功能基因组学中处于重要地位，并呼唤图像处理与统计数据分析之间的跨学科研究。研究人员将在这个界面上研究问题。即使是一个微阵列图像的数据量也是巨大的(30MB/扫描和两次扫描/图像)。压缩的需求迫在眉睫，因为它的使用呈指数级爆炸式增长，而且(尽管磁盘空间更便宜)到磁带的转换成本存在瓶颈，磁带是唯一可靠的永久存储。研究人员正在解决利用渐进或多级编码数据结构对微阵列图像进行压缩的这一需求，这对传输和统计分析是有用的。以有损压缩为例，在不损失太多统计信息的情况下获得15：1的积极压缩比。这项研究还研究了基于统计估计的有损压缩数据的最优化问题，并将其应用于微阵列数据。最后，本研究使用了基于数据压缩的统计建模原理：Rissanen的最小描述长度(MDL)原理，用于基因聚类等生物信息提取。

Bioinformatics is the application of information technology to study and analyze biological and genetic data. Functional genomics, the second phase of the Human Genome Project, assigns function to DNA sequences, and is gearing up rapidly based on the new microarray technology. Microarray images make possible the simultaneous expressions of thousands of genes, and hold the keys to understanding of gene regulation and interaction, genetic pathways for diseases such as cancer, and finding the subpopulations most responsive to certain drugs. They are widely used in laboratories of academia and industry alike. The emergence of the microarray imaging technology puts image processing in an important position in functional genomics, and calls for interdisciplinary research between image processing and statistical data analysis. The investigators will be studying problems at exactly this interface.The amount of data for even one microarray image is HUGE (30 MB/scan and two scans/image). The need for compression is pressing because of the exponential explosion of its use and (despite of the cheaper disk space) the bottleneck on the conversion cost to tapes, the only reliable permanent storage. The investigators are addressing this need for microarray image compression with a progressive or multi-level coded data structure which is useful for transmission and statistical analysis. A case is made for lossy compression to obtain aggressive compression ratios such as 15:1 without much loss of statistical information. This research is also investigating the optimality question of statistical estimation based lossy compressed data with applications to microarray data. Finally, this research is using the statistical modeling principle based on data compression: Rissanen's minimum description length (MDL) principle, for gene clustering and other biological information extraction.