Guiding Drug Selection for Ovarian Cancer through Machine Learning

Summary

We plan to develop a machine learning tool to optimize the chemotherapy of ovarian cancer, the most lethal gynaecological cancer. Many drugs are available for treating ovarian cancer, however, clinical outcomes vary substantially between patients. This suggests that a "personalized medicine" approach could be used to improve the choice of drugs and reduce ovarian cancer mortality rates.
Here, we will establish an entirely novel approach to guiding drug selection. We have full ethical approval to collect prospective tumour samples taken from individuals with suspected ovarian cancer at diagnosis or initial surgical treatment. These samples will be used to grow "cancer organoids": small tissues with similar genetics and structure to the parent tumour (doi.org/10.1038/s41591-019-0422-6). Large numbers of cancer organoids can be grown and used to screen panels of anticancer drugs. This has been touted as potential route to tailor anticancer therapy to individuals, however, we believe this is unrealistic due to the high cost and time required to produce cancer organoids for each patient.
In order to realise widespread clinical impact, we wish to develop a machine learning tool that will link organoid drug screening results to gene sequencing performed on the parent tumour biopsy. Once trained, this algorithm will be able to make intelligent predictions of the best-performing drugs based solely on the genetics of the tumour (i.e., without needing to grow organoids). This pump-priming funding would enable us to collect linked datasets of organoid drug screening and genetic sequencing, and then build and train a prototype machine learning tool.