A Bayesian Approach to Modeling Tamoxifen Resistance in Breast Cancer Cells Through Adaptive Hamiltonian Monte Carlo Posterior Sampling
Published in University of Deusto, Bilbao, Spain, 2025
This thesis is publicly available here.
Author: Miguel Fernandez-de-Retana
Supervisor: Aitor Almeida
Keywords:
Bayesian Computational Statistics \(\cdot\) Breast Cancer Research \(\cdot\) Gene Expression Analysis \(\cdot\) Hamiltonian Monte Carlo (HMC) \(\cdot\) Hormone Therapy \(\cdot\) Tamoxifen Resistance
Abstract:
Breast cancer is a major health concern globally, with diagnoses and projected cases rising significantly, becoming the most prevalent cancer among women and a leading cause of cancer-related deaths in the population. Hormone therapies, particularly tamoxifen, are vital for treating estrogen receptor-positive (ER$^+$) breast cancer, drastically improving survival rates by reducing recurrence and mortality. However, a significant challenge remains, as a substantial portion of patients develop resistance to the drug (estimated to be between 30-50%) within a critically long 5-year treatment window. To address this, our project proposes to analyze the complex mechanisms driving tamoxifen resistance. Employing a Bayesian modeling framework, and leveraging RNA sequencing data from cell-lines (in collaboration with the CIC bioGUNE research center) and publicly available patient data, the study aims to unravel the intricacies of this resistance phenomenon. To this end, we have further developed pyHaiCS, a Python library for Computational Statistics featuring a wide range of Hamiltonian sampling algorithms, including single-chain and multi-chain variants; a variety of numerical schemes for the integration of the simulated Hamiltonian dynamics, or a novel adaptive algorithm for the automatic tuning of the parameters. Finally, the project aims to identify key genetic biomarkers linked to tamoxifen resistance and develop robust, clinically applicable predictive models for patient prognosis under endocrine therapy. Ultimately, identifying therapeutic targets, paving the way for personalized and more effective treatments to improve patient outcomes. In practice, the methodologies developed here are intended to be broadly generalizable to other cancer types and drug resistance mechanisms.