Machine learning for functional reconstruction and analysis of biochemical interactions controlling colorectal cancer phenotypes

Abstract

Colorectal Cancer (CRC), broadly described as a cancer that affects the bowel and rectum, has a high prevalence, causing 10% of the cancer-related deaths in the UK. The 60% expected increase in deaths from CRC by 2030, which is already the fourth leading cause of cancer-related deaths in the world, raises a pressing need for better clinical tools for CRC patient stratification and treatment.Context-specific network inference (CoSNI) is a tool capable of integrating various functional genomics data to make predictions that capture context-specific molecular features to construct biochemical networks. Developed into an R package capable of handling cancer datasets using modified data pipelines generated in this project, it was applied to integrate polyomics data focused on CRC to capture molecular features specific to tumour stage and subtype. Networks for the following CRC contexts, stage II, stage III, CMS2 and CMS3, were successfully produced using CoSNI. These contexts were chosen for their influence on clinical decisions centred on prognosis and treatments.The four constructed networks were analysed using the Marisa dataset to interpret the network models using overall survival and response to chemotherapy to understand context-specific molecular mechanisms controlling these endpoints. I developed a workflow incorporating the network construction with the network analysis steps to provide insights underlying the CRC stages and subtypes.The CRC context-specific stage II network, when analysed, was able to identify four strong potential candidate biomarker genes driving five-year good and poor overall survival groups. Next, the CRC stage III network analysis identified the two independent genes as consistently good and poor prognostic markers respectively. This project significantly enhances CoSNIs' applicability to cancer data sets, making it a robust predictive tool to construct context-specific cancer biochemical interaction networks. Constructing four CRC-specific networks opens new avenues for precision medicine by offering personalised treatments.

Thesis embargoed until 31 December 2026.

Date of Award	Dec 2024
Original language	English
Awarding Institution	Queen's University Belfast
Supervisor	Ian Overton (Supervisor) & Sandra Van Schaeybroeck (Supervisor)