Autoregressive models for periodicity detection in DNA microarray time series data

Abstract

In a DNA microarray experiment, expression levels of thousands of genes are recorded simultaneously so that the functions of genes, the effects of certain therapies, disease, and developmental processes, among others, can be studied. With microarray technology, genome-wide gene expression data are being generated at a rapid rate. Biologists are interested in identifying the characteristics, trends, and patterns of gene expression profiles from a series of microarray experiments. However, each gene expression profile usually contains a certain amount of noise. It remains difficult to identify periodic gene expression profiles, especially when the number of data points is small and the level of noise is high. To increase accuracy when detecting periodic profiles, a noise filtering technique is needed before analysis of the gene expression data. We propose a new scheme combining singular value decomposition (SVD) with singular spectrum analysis (SSA). By considering the singular values of time series data, the trend component is extracted effectively so that noise can be filtered out. To detect the period of a time series, an autoregressive (AR) model-based estimation is used to assess the power spectrum density. Several datasets, including simulated sinusoidal signals and real DNA microarray time series experiments, are used to evaluate our scheme. The results show that our algorithm can reduce the noise level significantly and locate the correct period of time series signals.

Date of Award	3 Oct 2011
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Hong YAN (Supervisor)