Viktoria Samuelsson

Modelling and normalization for SELDI-TOF spectra

Abstract: Detection of cancer at an early stage results in more easily
treatable cases and ultimately in higher cure rates and less
treatment-related morbidities. One possible method for early cancer
detection is to use mass spectrometry, to find unusually high or low
concentrations of specific proteins in the blood samples from cancer
patients. Such proteins can then be used as biomarkers for diagnosis
and prognosis.  In this thesis we used mass spectrometry data from the
Oncology Department at Lund University. Our aim was to create a model
which can be used to separate the breast cancer patients from the
healthy patients. Since mass spectrometry involves many steps, we
decided to look closer at the variation in and between samples. We
also focused on the normalization process, by viewing the process of
intensity normalization as a regression problem. We used a
heteroscedastic model for normalization, which seems to have lower
variance then the usual normalization model.