Finding Regions of Amplification and Deletion in Array-CGH Data Using
Local Polynomial Regression and Change Detection Methods


Abstract: Genomic alterations in cells, such as gain and loss of
chromosomal material, are responsible for many severe diseases; among
them different kinds of cancer.  Knowledge about which genes are
involved is important when designing cancer therapy. A microarray
version of Comparative Genomic Hybridization, array-CGH, measures the
relative amount of chromosomal material in cancer cells and normal
cells at a large number of well known locations along the entire
genome simultaneously. The result of an array CGH experiment is a
series of ratios for each chromosome. The ratios are centered around
one for regions with no copy number aberrations, with peaks at
locations with gain and valleys at locations with loss of chromosomal
material. In this thesis, two methods for finding peaks and valleys in
array-CGH data are studied.  The first method, local polynomial
regression, is used to estimate where the function is significantly
increasing or decreasing. Especially, the SiZer tool for finding
significantly positive and negative derivatives is studied. The second
method, change detection, is used to identify abrupt changes of the
mean value in the data.