Research On Multiple Testing With Applications In Biological Data

Multiple hypotheses testing has a variety of applications in real life such as medicine, bioinformatics, genomics. brain imaging, etc.. And it focuses on finding more powerful procedures within the control of FWER or FDR. One key problem is the ratio of the true hypotheses in null ones.Several estimation methods are reviewed and a new method is proposed based on Hochberg&Benjamini [3] and Storey&Tibshirani [17]. We then com-pare our methods with the above two and Storey [18] through simulations. The comparison results suggest the general smaller bias by our method com-pared with the other three in the case of the number of hypotheses over2000. Finally we apply them in microarray data:the result also confirms the new method’s superiority in detecting biologically meaningful genes over the ones in Hochberg&Benjamini [3] and Storey&Tibshirani [17].Section one reviews roughly the basic framework of classical hypothesis testing theory. Section two is on the concepts of FDR and FWER and related procedures in multiple hypotheses testing. Section three introduces several estimation methods on the ratio of true hypotheses and proposes new method and compares them through simulations. Section four applies the estimation methods to microarray data and make comparisons.