Gene Expression Profile Analysis Based On Continuous Wavelet Transform

High through-put biological technology allows biologists to simultaneously monitorthe expression for hundreds of thousands of genes in organisms, and the resultingexpression data have become one of bases supporting modern biological genetics andmodern medicine. Currently, genome-wise research of cancer using gene expressionprofile data receives more and more concerns in bioinformatics community. Generally,gene expression data are highly dimensional, highly noisy, highly redundant, highlyvariable but with small sample, so that many traditional pattern recognition methodsand statistical methods are non-applicable or perform badly.By viewing gene expression profile as a"time-series"signal, this thesis proposesto apply continuous wavelet transform methods to analyze gene expression data. Es-pecially, we focus on gene expression pattern extraction and cancer classification. Aswe all know, wavelet transform can e?ciently decompose and reconstruct time-seriessignal and has been applied to various areas. Generally, wavelet transform can be im-plemented in two ways, i.e., discrete wavelet transform (DWT) and continuous wavelettransform (CWT). Researchers have applied DWT to gene expression data analysisand the results were good. In this thesis, in view of the better ?exibility in extractinginformation of CWT compared to DWT, we consider developing CWT-based methodsto analyze gene expression data. Because there are lots of useful information hiddenbehind gene expression data, using CWT to extract local details of gene expressionprofile is promising to perform better. In particular, we explore CWT-based methodsfor analyzing gene expression data in the following aspects: (1) How di?erent typesof wavelet base functions in?uence the performance of CWT in extracting expressionpatterns; (2) How varying values of the scale and translation parameters in?uence theperformance of CWT in extracting expression patterns;(3) How to integrate the well-known SNR gene selection method and CWT for better extracting gene expressionpatterns; (4) How the order of genes input to CWT in?uences the performance ofCWT in extracting gene expression patterns.Finally, on multiple publicly available cancer data sets, we tested the proposedCWT-based methods and extensively compared with previous methods. Experimentalresults show the better performance of CWT for gene expression data analysis com-pared with the previous methods. In summary, this work in this thesis has importantreference value for feature recognition and classification of tumors.