Penalized Pseudo-Likelihood Inference About the Dynamics of Multicellular Systems Using Flow Cytometry Data

Abstract

The potential of flow cytometry data in deciphering the dynamics of multicellular systems is tremendous but efficient statistical frameworks for their analysis are lacking. In this talk, we consider a stochastic finite mixture model defined using a Sevastyanov process allowing sharing of cell kinetic parameters across experimental conditions via covariate adjustment. A penalized pseudo-maximum likelihood estimator (PMLE) is proposed to take into account prior structural knowledge about fluorescence intensities. To compute parameter estimates efficiently we develop an EM algorithm, which considerably reduces computing times. We test hypotheses about cell kinetic parameters using a pseudo-likelihood ratio statistic designed to reduce the impact of optimization errors on the performance of the test due to the numerous nuisance parameters. Large sample properties of estimators are not always standard in the present context. For example, when cell counts are noisy, the PMLE convergences at a rate slower than $n^{-1/2}$ for some parameters, yet it is no less efficient than alternative estimators. Using the proposed approach we implement a stepwise procedure for comparing the kinetics of CD4+ and CD8+ T cells. The analysis suggests that fate determination and execution involve machineries that are not perfectly coupled. This is joint work with Rui Chen and Martin S. Zand.