Lamian: A Statistical Framework for Differential Pseudotime Analysis in Multiple Single-Cell RNA-Seq Samples

• Activity Number: 55 - Statistical methods for data from single cell technologies
• Type: Contributed
• Date/Time: Sunday, August 8, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Wenpin Hou* and Zhicheng Ji and Zeyu Chen and Stephanie C Hicks and Hongkai Ji
• Companies: Johns Hopkins Bloomberg School of Public Health and Duke University School of Medicine and Mass General Hospital/Harvard Medical School/Broad Institute and Johns Hopkins Bloomberg School of Public Health and Johns Hopkins Bloomberg School of Public Health
• Keywords: single-cell genomics; pseudotime analysis; multi-sample comparison; Bayesian hierarchical model; temporal gene expression; scRNA-seq

Abstract

Pseudotime analysis based on single-cell genomic data has been widely used to study gene regulation dynamics in continuous biological processes. However, methods that compare the pseudo-temporal patterns in one trajectory across multiple samples (e.g. donor, patient) of different conditions (e.g. disease severity, age) are lacking. Comparing pseudo-temporal gene expression in one trajectory between different sample groups is important to identify differential dynamic genes, cell type abundance, or other features that are associated with the sample covariates. We present a computational framework, Lamian, to (1)construct pseudotime trajectories from multiple samples and assess the uncertainties of the trajectories using bootstrap resampling techniques, (2) identify differential dynamic genes and cell distributions along a trajectory that are associated with a sample covariate using a Bayesian hierarchical model. Our method is able to control the false discovery rate and has higher statistical power than state-of-the-art methods. Using Lamian, we identified new molecular signatures in the comparison between COVID-19 mild and moderate patients during CD8 T cells activation.