High-dimensional Array Bayesian Screening Based on Distributions with Structural Zeroes

In many biomedical applications with high-dimensional features, such as single-cell RNA-sequencing, it is not uncommon to observe numerous structural zeros. Identifying important features from a pool of high-dimensional data for subsequent detailed analysis is often of interest. Here, we describe an exact, rapid Bayesian screening approach with attractive diagnostic properties, utilizing a Tweedie model. The method provides the likelihood that a feature with structural zeros merits further investigation, as well as distributions of the effect magnitudes and the proportion of features with the same expected responses under alternative conditions. The method is agnostic to assay, data type, and application. Through numerical studies, we demonstrate that the proposed methodology is effective in identifying important features for follow-up experimentation across a range of applications, including single-cell differential expression analysis of embryonic stem cells and embryonic fibroblasts in mice and differential analysis of CD4 and CD8 Peripheral Blood Mononuclear Cells (PBMCs) in humans.