Download

Download Full Text (417 KB)

Publication Date

5-2025

Abstract

Advancements in spatial proteomic imaging techniques have improved the ability to examine the tumor immune microenvironment (TIME) and determine its impact on clinical outcomes. Multiplex immunofluorescence (mIF) imaging is one such technique that can assess multiple markers simultaneously to differentiate the separate immune cell populations in the TIME. With the progress of immunotherapy (IO) treatments, analyzing these immune profiles has become increasingly important to determine tumors that may or may not respond to IO treatments. Despite the technological development, natural challenges arise when analyzing this data including the fact that the immune cell counts are over-dispersed and studies often have repeated measurements from the same tumor sample (i.e., cores on a tissue microarray, regions of interest). From a preliminary assessment of eight standard count-based distributions to model the spatial protein data, the beta-binomial distribution was the most interpretable, top fitting, model. Using a beta-binomial framework, we have developed a novel Bayesian hierarchical model to model multiple all immune cell populations of interest simultaneously. The Bayesian hierarchical model accounts for the relationships between the different cell populations in the TIME which is otherwise ignored from the standard univariate models. By incorporating different dependency (or relationship) structures between the different immune cell populations, there is improved precision in estimation due to “borrowing of information” across the multiple markers. Our Bayesian model and its different relationship structures for modeling the relationship between immune cell populations have been applied to spatial proteomic data from three large ovarian cancer epidemiologic cohorts (N = 486) to assess the ability of the novel model to detect factors associated with immune cell inflation into the TIME

Keywords

Bayesian, beta-binomial model, cancer, dependency, hierarchical, spatial protein imaging data, tumor immune microenvironment

Document Type

Poster

Bayesian Hierarchical Modeling that Leverages the Relationships Between Immune Cell Populations Measured from Spatial Proteomic Imaging Technologies

Share

COinS