CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Spatial survival models for analysis of exocytotic events on human beta-cells recorded by TIRF imaging Authors:  Thi Huong Phan - Department of Statistical Sciences, University of Padua (Italy) [presenting]
Abstract: In cell biology, exocytosis is a fundamental event observed in human beta-cells from high-resolution microscopy images. Studying the rate and spatial locations of exocytosis events, and predicting its survival probability, are of great interest in biomedical research as it helps to discover the cellular processes related to insulin-secretion dysfunctioning in diabetic patients. The main objective is to investigate the relationship between the exocytosis rate and syntaxin level observed during the experiments, while studying the possible spatial correlations within each cell. A Gaussian frailty survival model is proposed where individual spatial correlation is investigated through several different parametric families while independence clustering structure is preserved in the block pattern of frailty variance-covariance matrix. For estimation of model parameters, two common likelihood-based inferential approaches are firstly investigated: Monte-Carlo Expectation-Maximization (MCEM) approach and a penalized partial likelihood (PPL) approach. Finally, we propose a new approach where the marginal likelihood is estimated by pairwise likelihoods and quadrature approximation (QPLH). Their drawbacks and advantages are discussed in simulation studies, and also major results of the data application are presented, showing that exocytosis rates are spatially correlated and depend on their distance within each cell.