Date of Award
2008
Degree Type
Dissertation
Degree Name
Doctor of Philosophy
Program
Statistics and Actuarial Sciences
Supervisor
A.I. McLeod
Abstract
Three topics in the analysis of microarray genomic data are discussed and improved statistical methods are developed in each case. A statistical test with higher power is developed for detecting periodicity in microarray time series data. Periodicity in short series, with non-Fourier frequencies, is detected through a Pearson curve calibrated to the null distribution obtained by computer simulation. Unlike other traditional methods, this approach is applicable even in the presence of missing values or unequal time intervals. The usefulness of the new method is demonstrated on simulated series as well as actual microarray time series.
The second topic develops a new method for detection of changes in DNA or gene copy number. Regions for DNA copy number aberrations in chromosomal material are detected using maximum overlapping discrete wavelet transform (MODWT). It is shown how repeated application of MODWT to a series can be used to confirm the presence of change points. Application to simulated as well as array CGH (Comparative Genomic Hybridization) data confirms the excellent performance of this method. In the third topic, it is shown that an improved class predictor for tissue samples in microarray experiments is developed by incorporating nearest neighbour covariates (NNC). It is demonstrated that this method reduces the mis-classification errors in both simulated and actual microarray data.
Recommended Citation
Islam, Mohammad Shahidul, "Peridocity, Change Detection and Prediction in Microarrays" (2008). Digitized Theses. 3210.
https://ir.lib.uwo.ca/digitizedtheses/3210
Comments
At the request of A.I. McLeod