B0878
Title: Supervised heterogeneous network estimation via survival-based Bayesian graphical models
Authors: Xing Qin - Shanghai University of International Business and Economics (China) [presenting]
Shuangge Ma - Yale University (United States)
Mengyun Wu - Shanghai University of Finance and Economics (China)
Abstract: Reconstructing biological networks from high-dimensional gene expression data remains an important task in systematically understanding the disease mechanisms. Recent studies often explore network structures in an unsupervised learning paradigm without considering any information about the clinical subtypes of patients. Although fewer studies investigate supervised network learning in low-dimensional settings, they are not scalable to high-dimensional settings and fail to identify both common and varying substructures across subtype-specific networks. To deal with the joint estimation of multiple large networks accounting for the unknown clinical-relevant disease subtypes, a novel supervised heterogeneous network estimation approach is developed via survival-based Bayesian graphical models. It is among the first supervised methods that conduct joint estimation for multiple networks with unknown subtype structures. The approach combines Gaussian mixture models with accelerated failure time models to significantly facilitate clinically meaningful biological network construction while accommodating similarities among patients with different subtypes. Theoretically, the obtained estimators achieve consistent properties. Extensive simulation studies and an application to TCGA data are conducted, which demonstrate the advantages of the proposed approach in terms of both subtype and network identification.
