A0454
Title: Identification of differentially expressed genes via knockoff statistics in single-cell RNA sequencing data analysis
Authors: Lixia Yi - University of Pittsburgh (United States)
Linxi Liu - University of Pittsburgh (United States) [presenting]
Abstract: Single-cell RNA sequencing (scRNA-seq) is a technology that provides high-resolution gene expression data. With scRNA-seq data, an important statistical task is to identify differentially expressed genes (DEGs) in case-control studies, as results from DEG analysis can contribute to a more comprehensive understanding of the disease mechanism and new discovery of potential risk factors. However, given the burden of multiple testing and low transcript capture rate in scRNA-seq experiments, DEG identification may suffer from low power. Co-expressed genes and unobserved confounders can also lead to an inflated Type-I error. A new method for DEG identification is introduced in scRNA-seq data analysis under the knockoff framework to overcome these difficulties. The method starts by imputing missing gene expressions by taking advantage of correlations among genes, and then it generates model-X knockoffs in a computationally efficient way. By incorporating widely used marginal screening tests for scRNA-seq data, we implement a knockoff filter for DEG identification that can control the false discovery rate (FDR) at the nominal level. On a range of synthetic and real data sets, FDR control and power gain of the new approach are illustrated. The method is also applied to the single-cell transcriptomic analysis of Alzheimer's disease. The results demonstrate that the new method can identify genes with weaker effects that are missed by conventional approaches.
