Using latent process decomposition to classify prostate and colorectal cancers.

Ellis, Christopher (2021) Using latent process decomposition to classify prostate and colorectal cancers. Doctoral thesis, University of East Anglia.

[img] PDF
Restricted to Repository staff only until 31 August 2022.

Download (9MB) | Request a copy

Abstract

Cancer classification plays an important role in the clinical management of cancer patients. It enables clinicians to predict how individual cancers will behave and directs the best course of treatment. However, the classification of heterogeneous cancers has proven to be challenging.

To address this problem more advanced classification techniques should be used. In this thesis we focus on the unsupervised Bayesian algorithm Latent Process Decomposition (LPD). This technique has previously been used to classify breast cancer and was recently used to produce a novel classification of prostate cancer. We therefore aim to leverage LPD’s ability to classify heterogeneous diseases.

We begin by performing a study on the prostate cancer subtype DESNT, introduced by Luca et al. (2017). By creating and applying a new type of LPD algorithm (OAS-LPD) to the DESNT classification, we establish a DESNT risk score that is an independent predictor of progression alongside existing diagnostic variables (PSA level and Gleason score). DESNT’s expression profile is also demonstrated to be detectable in prostate cancer biopsies. Combined, these findings present the possibility for a new clinical test to reduce the over treatment of prostate cancer patients.

In the second part of this thesis we apply LPD to six transcriptome datasets obtained from colorectal cancer (CRC) biopsies. We identify and characterise four new CRC subtypes present across the datasets, including one subtype (designated Pericol) associated with a statistically significant poorer prognosis. Many of the Pericol signature genes are shown to overlap with other published signatures and the Pericol risk score is identified as an independent predictor of disease recurrence.

Our results demonstrate the existence of poor prognosis categories of human cancers that can be used to assist in the targeting of treatment. They also emphasise the importance of employing biologically appropriate techniques to classify heterogeneous diseases.

Item Type: Thesis (Doctoral)
Faculty \ School: Faculty of Science > School of Computing Sciences
Depositing User: Chris White
Date Deposited: 26 Aug 2021 07:53
Last Modified: 26 Aug 2021 07:53
URI: https://ueaeprints.uea.ac.uk/id/eprint/81239
DOI:

Actions (login required)

View Item View Item