Susann Stjernqvist
	
Analysis of CGH-data with hidden Markov models

Abstract: Array comparative genomic hybridization (CGH) provides
information about the number of copies of DNA clones. This is done by
labelling a test sample with a fluorescent dye and a reference sample
with another fluorescent dye, and then comparing the intensities when
radiation with a laser beam. By this we can find regions with gains
and losses of DNA. The aim of this thesis is to create a method by
which aberrant regions can be found. We have studied two different
methods, which both use hidden Markov models. One method is in
discrete time, and then we use the EM algorithm to find the hidden
states for each clone. The clones sometimes overlap, and to use all
that information we work in continuous time in the second model. Then
we use a Bayesian approach and MCMC to find the aberrant regions. Our
expectations was that the continuous time model would reduce the
correlation between the residuals, but unfortunately there was not a
general reduction. The discrete time method has the advantage that it
has much shorter execution time. We have also studied an expansion of
the models from having a common variance for all states to different
variances for different states.