Tracking the evolution of cancer genomes

Abstract:: Tumour clones consist of cancerous cells with shared ancestry. The study of cancer evolution has been further advanced by the introduction of mathematical algorithms to reconstruct the phylogenetic tree within the tumour. Bulk sequencing data pro- vides not only the VAF and ploidy information that most established methods now use to infer tumour subclonal structure, but also another equally important piece of information, namely phasing information. Phasing information from sequencing reads can be used to find the most likely phylogenetic trees.

During my Ph.D, I proposed a generative model of SNVs distribution in next-gen sequencing reads, PhaDPClust, which utilises phasing information and constructs complete phylogenetic trees automatically. This method builds on the DPClust framework, and can be applied to both single sample and multiple sample analysis.

On single sample analysis, PhaDPClust outperformed other established methods in subclone reconstruction, measured by several metrics from the SMC-HET Chal- lenge. The performance of PhaDPClust was constrained by the abundance of phase information, as evidenced by results on real tumour samples.

Then, PhaDPClust was extended to multisample analysis. PhaDPClust suc- ceeded in reconstructing phylogenetic trees in similar topology automatically, but failed to split some clusters, implying some work still needs to be done in the future.