Computational approaches and their applications in cancer: uncovering the role of extracellular vesicles, hypoxia and cancer phenotypes in the tumor microenvironment

Cancer phenotypes typically arise from 1) genetic and epigenetic deregulations and 2) altered signaling responses, a consequence of aberrant interactions within the tumor microenvironment (TME). The study of the TME and, broadly speaking, of the tumor complexity is well suited to quantitative approaches, and provides opportunities for methodological developments. In turn, in silico models and integrated analyses are instrumental to predictively model cancer behavior, with the aim of dissecting key mechanisms of tumor initiation, progression, dissemination and drug resistance. The main objectives of this dissertation involve the development and the application of computational methods to disentangle cancer complexity. Several research questions have been addressed, ranging from basic research to translational applications. A data-driven investigation of extracellular vesicles (EVs)-mediated cell signaling was conducted by characterizing the molecular cargo of tumor-derived EVs released by Neuroblastoma (NB) cells. Predictions were confirmed by in vitro and in vivo validation and revealed a microRNA-signature associated to tumor aggressiveness in hypoxic microenvironments. As a complementary approach, the development of a first-principles model to quantitatively characterize EVs diffusion and uptake was proposed to study the short- and long-range effects of released EVs. A comparative study on feature selection and molecular classification of cancer phenotypes allowed evaluating the impact of algorithmic combinations to predict cancer phenotypes using gene expression profiles. A set of order effects for successful classification of cancer phenotypes stemmed from this study, which constitutes a valuable resource in the view of designing diagnostic and prognostic tools. However, the elucidation of cancer-associated dysregulations that coordinately shape malignant cell states cannot be uniquely based on reductionist approaches, but requires systems biology. In this work, network-based analysis of genomic data – especially single-cell RNA sequencing (scRNA-Seq) data – made it possible to dissect tumor heterogeneity at single-cell resolution, in the context of several malignancies, including prostate and pancreatic cancer. The analysis of scRNA-Seq poses several computational challenges that require the development of suitable methods. To this aim, an optimization-based framework for graph-based clustering of cell populations and a computational framework to dissect ligand/receptor-mediated paracrine crosstalk between distinct cellular niches are proposed in this dissertation. Moving to the task of performing phenotypic characterization of in vitro tumor spheroids, an image-analysis based pipeline was developed to quantify cytotoxic effects of mycotoxins on NB cells. This contribution constitutes a valuable tool for the extraction of quantitative features from spheroids images, and was key for the evaluation of several effects such as morphological shifts and decrease in migration following mycotoxins exposure on tumor cells. While remarkable progress has been made towards quantitative cancer research, multiple open questions remain, including many related to understanding the dynamics that take place in the TME. Approaches and results presented and discussed in this dissertation make several steps in pushing research forward, towards a more quantitative understanding of the mechanisms that nurture cancer, with the aim of allowing deeper insights into the complexity of biological systems.