Characterizing heterogeneity between cancer patients by integrating molecular data, imaging data and pre-existing knowledge

Abstract

Because every cancer is different, different cancer patients can benefit from different treatments. Therefore, to give every patient the best treatment possible, the differences between cancers need to be described. In this thesis, we describe this heterogeneity in several contexts: in breast cancer, in a subtype of breast cancer, invasive lobular carcinoma, in lung cancer and across multiple cancers types. To this end, we employ multiple data types, including magnetic resonance images (MRI), mutations, copy number aberrations, mRNA expression, and protein expression, as well as existing biological knowledge. The first chapter introduces this thesis by describing the different data types available and by classifying the methods for describing heterogeneity. Methods which employ early data integration are distinguished from methods which employ late data integration, and similarly for early and late knowledge integration. In the second chapter, a multilevel hierarchy based on existing knowledge is built on which mutation and copy number data are mapped to describe co-occurrence and mutual exclusivity. In the third chapter, mutation, copy number aberration, mRNA expression, and protein expression data are integrated which leads to the identification of two subtypes in invasive lobular carcinoma that could potentially benefit from different treatments. In the fourth chapter, we introduce a new method for data integration, functional sparse-factor analysis. It describes the heterogeneity by identifying continuous factors along which the data varies and provides a functional interpretation of these factors. Its applicability is shown on breast and lung cancer. In the fifth chapter, we calculate a measure of estrogen receptor pathway activity and relate this to an MRI feature: contralateral parenchymal enhancement. In the sixth chapter, we link gene expression data to MRI features to improve the biological understanding of these features. The thesis ends with a chapter that discusses what have learned about methods to integrate existing knowledge and multiple data types, the insights gained about the heterogeneity in cancer from applying these methods and draws some future perspectives.