Tackling the development of hormone therapy resistance in breast cancer through mathematical modelling

Abstract

Patients suffering from estrogen-driven breast cancer frequently develop hardly predictable resistanceto hormone therapy, which creates a significant clinical challenge. Current approaches for tackling thisproblem include cell models and clinical studies, both supported by sequencing technologies likeRNA-seq, and offering different strengths and limitations. This dissertation addresses the challenge ofpredicting resistance to hormone therapy in breast cancer by merging advances in bioinformatics andBayesian statistics, and applying them to two types of data ¿ RNA-seq data and clinical data. First, weexplore the statistical analysis of clinical data through Bayesian inference combined with enhancedMarkov Chain Monte Carlo techniques, and introduce a novel algorithm for adaptive integration inprospective Modified Hamiltonian Monte Carlo (MHMC) methods. We demonstrate its positive effecton performance of MHMC in biomedical applications using clinical data of breast cancer patients.Next, we propose and implement an RNA-seq pipeline within our interactive web-app for the analysisof resistant breast cancer cell lines sequenced at CIC bioGUNE. Finally, we propose an originalapproach based on a Bayesian logistics regression model coupled with a simulated annealing-likealgorithm for a combined analysis of RNA-seq and clinical data, and apply it to ad hoc data to obtainand validate in-silico and in-vitro a novel 6-gene signature for stratifying patient response to hormonetherapy.