Year: 2019
Type: dissertation
Abstract: Tumours are composed of multiple subpopulations, each of which has its own genotype and phenotype. These genotypes are related to each other in an evolutionary tree, where the root of the tree is the germline genome. Subpopulation genotypes are therefore typically characterized in terms of their somatic mutations. These mutations are responsible for the tumour's properties . Finding the genotype of these different subpopulations can help study cancer biology and provide more accurate predictions of patient outcomes. Determining the genotype of these subpopulations from a limited number of sequenced bulk tumour samples is the subclonal reconstruction problem. As the number of subpopulations present is not known, and the subpopulations are related to each other in a tree, using non-parametric, tree-structured clustering algorithms is a natural approach to solve this problem. I present three advances in subclonal reconstruction. In the first I present an algorithm that expands the set of mutations available for subclonal reconstruction by modeling the interaction between two types of mutations, Single Nucleotide Variants (SNVs) and Copy Number Alterations (CNAs). CNAs affect the observed evidence of SNVs in a way that depends on their evolutionary relationship. Thus, to perform subclonal reconstruction on CNA-affected SNVs, the reconstruction must include both types of mutations. I develop the method PhyloWGS and demonstrate accurate subclonal reconstruction in the face of SNV/CNA interactions. The choice of non-parametric prior used affects subclonal inference. Different priors favour different trees and have different technical attributes. I compare the treeCRP prior against the Tree Structured Stick Breaking (TSSB) prior. This is the first use of the treeCRP prior in subclonal reconstruction. I found that the treeCRP had better mixing and lower computational cost compared to the TSSB but generated less accurate subclonal reconstructions. When performing subclonal reconstruction multiple trees can fit the observed data equally well. Incorporating pairwise haplotypes in subclonal reconstruction reduces this ambiguity. These haplotypes are inferred on the basis of reads that span the loci of two mutations. I incorporate pairwise haplotypes in subclonal reconstruction with a new algorithm called PhyloSpan. I demonstrate more accurate subclonal reconstruction with reduced ambiguity using PhyloSpan.
Source: TSpace (University of Toronto)
Author: Amit G. Deshwar
Language: English
Subfield: Cancer Research
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W2963167328