| Many preprocessing methods and differentially expressed gene selection methods have been proposed since the appearance of high-density oligonucleotide arrays. Most of those methods are based on some assumptions about the data. However, in most situations, the assumptions do not hold; therefore all the methods have some limitations due to their unreasonable assumptions. In this dissertation, many current popular preprocessing methods are investigated and some of their limitations are indicated. To overcome their shortcomings, some new parameter estimation methods for certain models have been proposed; new non-parametric methods dealing with background correction and summarization are also proposed.; For identifying differentially expressed genes, a novel method based on probe-level data is also proposed. Compared with current methods, this method has more power; it also avoids the so-called multiple comparison problem. |
