SifiNet: A robust and accurate method to identify feature gene sets and annotate cells

Abstract

Single-cell sequencing has provided a means of quantifying cellular omic phenotypes. Identifying cell-type-specific feature genes is a crucial aspect of understanding cellular heterogeneity. Over the past decade, many methods have been developed to identify feature genes; however, these methods either depend on dubious cell clustering or fail to provide subpopulation-specific markers. We introduce SifiNet, a robust and accurate approach for identifying marker gene sets based on gene co-expression network topology. The identified gene sets facilitate the calculation of cellular gene set enrichment scores and cell annotation, and can reveal potential transitional relationships between cell subpopulations. SifiNet outperforms state-of-the-art methods in marker gene set identification and cell-type annotation accuracy. It is applicable to both single-cell RNA sequencing (scRNA-seq) and single-cell ATAC sequencing (scATAC-seq) data. We have applied SifiNet to various experimental studies, successfully identifying novel gene markers, annotating cells with complex heterogeneity, and uncovering intriguing cell developmental trajectories.

eSVD-DE: cohort-wide differential expression in single-cell RNA-seq data using exponential-family embeddings

“…We use the multiple testing procedure based on the empirical null developed in [57] to handle the multiple testing corrections. This framework is also used in other methods designed to test for differential expression from scRNA-seq data (albeit not from cohort data), such as iDEA [58] and SifiNet [59]. We know that biologically, most genes are not directly related to the disease or disorder at the human population level; hence, most genes should be deemed insignificant.…”