Hierarchical Analysis of Time-Course Flow Cytometry Data with Dirichlet Process Mixture Modeling Applied to HIV Vaccine Trial Data

• Activity Number: 36
• Type: Contributed
• Date/Time: Sunday, August 3, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Boris Hejblum*+ and François Caron and Rodolphe Thiébaut
• Companies: INSERM/INRIA and University of Oxford and INSERM/INRIA
• Keywords: Dirichlet Process Mixture ; Flow-cytometry ; Longitudinal Data

Abstract

Flow cytometry is a high-throughput technology used to quantify multiple surface and intracellular markers at the level of a single cell. Improvements of this technology lead today to the ability of describing millions of individual cells from a blood sample, across multiple markers. This results in huge datasets, whose manual analysis is highly time-consuming and poorly reproducible. Several methods have been developed to perform automatic recognition of cell populations from flow cytometry data. But most of them are based on single sample in one patient, typically using 4 colors (dimensions) only. In clinical trials, repeated measurements with several samples by patient and by time points are actually available. We propose to use a Bayesian nonparametric approach with Dirichlet process mixtures (DPM) to model such data. DPMs enable the number of cell populations to be estimated from the data, without resorting to model selection. We apply this approach to the DALIA-1 trial, a clinical trial evaluating a therapeutic vaccine against HIV. Flow cytometry measurements, with several samples using 6 to 8 colors at each of the 18 time points, were performed for each of the 18 patients.