(744b) Bayesian Inference Based Aggregated Gaussian Process Models for Identification of Ovarian Cancer Subtypes and Prediction of Survival Rates of Cancer Therapies

Cancer genome research effort aims to provide a comprehensive molecular characterization to accelerate the understanding of cancer biology and the discovery of new therapeutic targets. The immense wealth of multidimensional genomic data provides a new paradigm for developing new therapeutic solutions to improve the lives of patients.

The standard treatment for high-grade serous ovarian cancer is aggressive surgery followed by platinum-taxane chemotherapy. After therapy, platinum resistant cancer can occasionally reoccur resulting in a low overall five-year survival rate. The comprehensive measurements of genomic and epigenomic abnormalities on high-grade serous ovarian cancer samples are essential to identify the molecular abnormalities that may affect the outcome of different treatments. Moreover, accurate identification of the ovarian cancer subtypes can provide an opportunity to improve specialized cancer therapies. To that end, a genomic identification method based on Bayesian inference and Gaussian process classification is developed by using the gene expression data to identify the subtypes of cancer and predict the survival rates of cancer therapies. The proposed approach uses aggregate models developed through the integration of local Gaussian process classification models using the posterior probabilities estimated from Bayesian inference. Unlike the current standard analysis that involves separate clustering of different genomic data types followed by a manual integration of the cluster assignments, the new approach results in various localized models that can be amalgamated within the Bayesian framework. The application of the proposed approach to ovarian cancer patient samples demonstrates the significantly improved prediction accuracy and identification performance over the traditional techniques.