Exploring disentangled generative factors of cancer transcriptomes

Cancer is a complex disease with genetic and environmental causes. The treatment of cancer has enormous complexity because molecular-level subtypes of the disease are not always apparent as phenotypes at the pathological levels, thus requiring stratification of patients, even down to individuals. The inability to do this results in any one of the vast array of oncological drugs available to clinicians eliciting a positive response only on a small fraction of sufferers of the disease.
With recent advances in high throughput experimentation, large number of measurements at the level of genome aberration, nucleotide mutations, gene expressions, protein concentrations and regulatory information is acquired. It is accepted that our ability to acquire data outstrips our ability to make accurate and useful inferences from them.

This project will explore the development and application of state-of-the art unsupervised learning methods to help understand the mapping between genotypic measurements and cancer phenotypes. Generative models that capture properties of the intricate tumour micro-environment will be developed. We will aim to disentangle the contributions of various factors influencing tumors, their progression and response to treatment. The work will be based on a rich repository of tumor related database: The Cancer Genome Atlas.