A0294
Title: Latent modularity in multi-view data
Authors: Xuejun Yu - National University of Singapore (Singapore) [presenting]
Andrea Cremaschi - ASTAR (Singapore)
Maria De Iorio - National University of Singapore (Singapore)
Ajay Jasra - KAUST (Saudi Arabia)
Shu Qin Delicia Ooi - National University of Singapore (Singapore)
Xiu Ling Evelyn Loo - ASTAR (Singapore)
Navin Michael - ASTAR (Singapore)
Abstract: Asthma, hypertension and obesity are three of the most common chronic diseases worldwide, with known presence of comorbid pathophysiological mechanisms. Despite studies indicating a complex coregulatory mechanism between these diseases exists, quantitative analyses in children are currently scarce. Furthermore, data collected from different sources are usually analyzed separately, neglecting shared information among subjects, underlining the need for a more comprehensive statistical approach. A novel Bayesian nonparametric model is developed for the joint analysis of biomarkers of different types related to obesity (longitudinal data), history of asthma (panel count data) and symptoms of hypertension (multistate process). Random partitions of subjects in each dataset are modelled independently conditionally on an underlying partition structure. The proposed strategy allows for sharing information among clustering structures within different datasets, thus providing more robust inference. Random partitions of different datasets are marginally dependent, with the level of dependence learnt from the data. The model involves mixed-type covariates, aiding the identification of risk factors affecting the evolution of diseases. A tailored MCMC algorithm is developed, which entails simpler computations than existing methods based on hierarchical random measures. An application from the Singaporean birth cohort GUSTO (Growing Up in Singapore Towards Healthy Outcomes) is presented.
