Using machine learning to predict subclone evolution and response during chemotherapy
This research aims to apply artificial intelligence (AI) approaches to genomic and clinical data to develop models that predict response to chemotherapy in breast cancer. AI and genomics are two technologies with the potential to transform cancer care. They are driving precision oncology, allowing personalized treatment focusing on identifying what works for an individual and avoiding what does not.
Chemotherapy is a mainstay of the treatment of metastatic breast cancer, that is breast cancer that has spread within the body. The response to chemotherapy is very variable. For some people treatment works very well, at least initially. For others, treatment has little or no effect. Being able to predict who will respond would allow treatment to be targeted to those most likely to benefit. Those unlikely to respond could be offered alternative treatments and spared unnecessary toxicities.
A complex interaction between drugs, cancer cells and the patient determines the response to chemotherapy. Much of a cancer’s behaviour is driven by mutations in genes controlling cellular functions, such as growth and division. The patterns of mutation can vary from one area of cancer to another and can change over time. As a result, genetically distinct subpopulations of cancer cells (subclones) can arise, which vary in their sensitivity to chemotherapy. Our approach to capture the genetic profiles of subclones is to analyse circulating tumour DNA (ctDNA). This is DNA that has been shed from the cancer into the bloodstream and which can be extracted from a blood sample.
Understanding the complex interactions that determine response to chemotherapy requires sophisticated data analysis techniques. Machine learning is a branch of AI that provides the ability to learn patterns from data. In this project, we will apply cutting edge machine learning techniques to an integrated ctDNA and clinical data set. Our models will predict which subclones will become predominant. This would allow treatment to be modified to target these subclones. Our models will also predict how well a patient will respond, allowing therapy to be individualized.
Insights from our public partners have guided both the strategic direction of our research plans, and specific issues relating to this project. A public representative is a co-applicant, enabling ongoing public partner interaction.
We will disseminate results by submitting papers to peer reviewed journals and presenting at conferences. We will seek to engage with service users, social care planners, practitioners and policy makers. Our study can be seen as an exemplar for integration of clinical and genomics data. It is likely, in the near future, that all cancer patients in Wales will receive comprehensive genomic profiling of their cancer as part of routine NHS care. By applying artificial intelligence approaches to this large-scale genomic and clinical data, we could further refine our models and deploy them into clinical practice. These approaches could extended to other treatments and cancer types. This could improve outcomes, avoid unnecessary toxicities and make effective use of high-cost drugs, leading to better quality and value cancer care within Wales.