Mining Local Linear Patterns From Microarray Gene Expression Data

The development of microarray technology makes it possible to measure thousands of gene expression rapidly and precisely at the same time. Therefore, how to mine valuable patterns from plenty of gene expression data has become one hot spot issue in Bio-infomatics recently. Pattern mining can help researchers to analyze gene expression data and find the hidden corelationships among genes. Hence, it can help human beings get to know genes and make use of genes.The existing methods of mining patterns mainly focus on mining the simple same-trend-patterns. However, the complex patterns, such as time-lag-patterns and variable-rate-patterns, are far more valuable in research and far more difficult to mine. Some new methods mining complex patterns are proposed recent years. However, some of them rely on the formulas calculating correlation degree. As such, they have low accuracy and it is hard to extend their applications. Some algorithms are inefficient due to computational complexity. More importantly, existing methods usually divide data which should have been put into one linear pattern into several sub-patterns. Hence, it is impossible to analyze these data entirely, leading to the missing of some interesting rules.In this thesis, we firstly have made a thorough investigation of existing pattern mining algorithms. Then we explore the relationship among various local patterns and define their mathematical models. Based on these models, we introduce a new categorization of local patterns and propose the notion of"linear pattern". We also analyze the pattern mining algorithms based on matrics transformation, induce their mathematical models and discover the relationship between these algorithms and derivative. We propose a novel model and algorithm to mine linear patterns. We prove the correctness of our model theoretically and implement our system for experiments.Our contributions are summarized as follows:1. Introducing a novel categorization of local patterns;2. Proposing the notion of"linear pattern";3. Connecting the matrics transformation with derivative; 4. Proposing a novel model and algorithm SDC to mine linear patterns.The algorithm SDC proposed in our research can be applied to mine linear patterns not only for microarray gene expression data but also for data in other fields such as commercial field and economical field.