A0484
Title: Regression analysis for single-cell RNA-seq data
Authors: Fangda Song - The Chinese University of Hong Kong, Shenzhen (China) [presenting]
Kevin Y Yip - Sanford Burnham Prebys Medical Discovery Institute (United States)
Yingying Wei - The Chinese University of Hong Kong (Hong Kong)
Abstract: scRNA-seq studies that assay a large cohort of donors are emerging, which provides opportunities for us to understand how gene expression profiles of a cell type are affected by donors' characteristics, such as age, gender and disease status. However, statistical methods developed to study the association between bulk gene expression data and a set of covariates are not applicable to scRNA-seq data because the cell-type label of each cell is unknown. In addition, batch effects, variations in cell-type abundance between donors, and missing data due to dropout events all add challenges to detecting the association between scRNA-seq data and covariates. Regress-seq, a Bayesian hierarchical model, is developed that can simultaneously cluster cell types, correct batch effects and the effects of the covariates inferred on cell-types-specific gene expression profiles. Moreover, the conditions are derived from the experimental designs under which the integrative analysis of multiple scRNA-seq studies is valid so that the cell type effects, batch effects and covariate effects can be separated. Regress-seq is envisioned to greatly facilitate the development of personalized medicine.
