B1958
Title: A robust kernel machine framework for assessing differential expression of multi sampled single cell data
Authors: Tusharkanti Ghosh - University of Colorado (United States) [presenting]
Abstract: CytoKernel is introduced, a robust method for differential expression analysis of single-cell data. Specifically designed for single-cell RNA sequencing and high-dimensional flow or mass cytometry data, this method leverages the full distributions. While high-throughput sequencing of single-cell data offers a detailed view of cell specification, many existing methods only focus on aggregate measurements, capturing only global changes. Unlike these, cytoKernel is built on a semi-parametric logistic regression model, utilizing the full distributions of single-cell data. It calculates the divergence between pairwise distributions of subjects, enabling detection of both aggregate changes and nuanced variations. These subtle changes are often missed due to the multimodal nature of single-cell data. We benchmarked cytoKernel using simulated and real datasets from single-cell mass cytometry and RNA sequencing. Our results indicate that cytoKernel effectively manages the False Discovery Rate (FDR) and outperforms existing methods in identifying differential patterns. We further applied it to evaluate gene and protein marker expression differences in various single-cell datasets.
