Detecting Novel Subtypes of Cancer Using Bayesian Unsupervised Clustering

Llaneza Lago, Sergio (2023) Detecting Novel Subtypes of Cancer Using Bayesian Unsupervised Clustering. Doctoral thesis, University of East Anglia.

Preview

PDF Download (10MB) | Preview

Abstract

Although there have been many advances in screening programs and treatments in recent years that have reduced the mortality rate of cancer, it remains the second leading cause of death worldwide, accounting for almost 10 million deaths worldwide in 2020. Identifying and characterising subtypes based on molecular classifications can help identify the aggressiveness of the disease so that the best treatment pathway can be identified, and new treatment options developed. This has been exemplified in breast cancer. Latent Process Decomposition (LPD) is a soft clustering technique that has been successfully applied to expression data to discover subtypes, including a poor prognosis subtype called DESNT. The benefit of LPD is that it better models the heterogenous structure of tumours.

The aim of this thesis is to apply LPD on transcriptome data from The Cancer Genome Atlas to detect and characterise subtypes of numerous cancer types and create a resource of the results. This was achieved through the development of Automata, an R package used to automate this methodology.

In total I have identified 168 cancer subtypes spanning across 28 cancer types. Moreover, I have characterised the features of each subtype, generating a unique encyclopaedic compendium of molecular subtypes of cancer that provides an in-depth source of information for the research community. I have successfully validated my findings by comparing them with known subtypes from breast carcinoma, prostate adenocarcinoma, colorectal adenocarcinoma and lung cancer. Additionally, I have discovered common features that characterise subtypes across cancer types. Finally, I have identified 26 subtypes which have a significant association with outcome including some that were not picked up by traditional clustering methods.

The results presented in this thesis are the foundation for the long-term impact of a more personalised approach to cancer patient care.

Item Type:	Thesis (Doctoral)
Faculty \ School:	Faculty of Medicine and Health Sciences > Norwich Medical School
Depositing User:	Chris White
Date Deposited:	12 Oct 2023 08:18
Last Modified:	12 Oct 2023 08:18
URI:	https://ueaeprints.uea.ac.uk/id/eprint/93273
DOI:

Downloads

Actions (login required)

View Item