Grant: DRA : Canada Graduate Research Scholarship - Doctoral Program (CGRS D)

Cancer is an evolutionary process. A key consequence of this is that most tumours are composed of multiple genetically distinct populations of malignant cells called clones. These clones play a major role in patient outcome. Thus, approaches which can precisely quantify the composition of clones in a tumour are critical. The ability to resolve tumour heterogeneity has improved with the development of high throughput single cell whole genome sequencing (scWGS) as well as other sequencing methods. However, due to low sequencing coverage these methods generally generate sparse and noisy data. We propose three projects that employ bayesian machine learning to infer clonal copy number profiles from noisy data. The first method will infer allele specific clonal copy number profiles from scWGS data. Using a mixture of Hidden Markov Models, this method will simultaneously infer the grouping of cells and clonal copy number profiles. By combining the steps of cell aggregation and copy number inference, this approach will share statistical strength to overcome noisy measurements. Our second project will be an extension of the first. This project will perform clonal inference from multiple data modalities using the mixture model based framework from the first project. We hypothesize that the additional information gained from other data modalities will improve the clonal inference. The third project will aim to infer clonal copy number profiles from spatial data. Few methods exist yet that can successfully infer clonal copy number profiles from spatial data. We hypothesize that by applying a bayesian framework to spatial data will result in effective spatial clonal inference and allow for a spatial analysis of the clonal landscape of a sample.