Classification of metastases based on cell type hererogeneity

Specialeforsvar ved Jing Zhou

Titel: Classification of metastases based on cell type heterogeneity

Abstract: Biological samples of actual interest sometimes contain a contamination in the form of certain well-specified additional sample materials. A biopsy of cancer tissue is often contaminated by the surrounding benign tissues such as liver, blood, etc.. Contamination of a cancer tissue is a concern for cancer diagnostics. In this project, we focused on the example- the molecular identification primary tumor site of metastases, see Vincent et al.(2014); since biopsies of metastases often contain a significant proportion of benign tissues. We were interested in a correct classification of the metastases biopsies under cell type heterogeneity (or contamination of the sample). The actual deconvolution was of secondary interest, as well as the proportion of the benign tissues. This project was based on the deconvolution model in linear space, which had been used in the previous studies (see Shen-Orr et al., 2010; Zhong and Liu, 2012). To carry out the classification under cell type heterogeneity, we constructed two Monte Carlo classifiers which are functions of the contaminated tissues. The two classifiers were denoted as the ”Estimated Monte Carlo classifier” and the ”Class-wise Monte Carlo classifier”. The actual deconvolution was completed using the Markov chain Monte Carlo method 

Vejleder: Niels Richard Hansen
Censor:   Per Bruun Brockhoff, DTU