Cofea: correlation-based feature selection for single-cell chromatin accessibility data

Abstract

Single-cell sequencing technologies have revolutionized the understanding of cellular heterogeneity at an unprecedented resolution. However, the high-noise and high-dimensional nature of single-cell data poses challenges for downstream analysis, and thus increases the demand for selecting biologically informative features when processing and analyzing single-cell data. Such approaches are mature for single-cell RNA sequencing (scRNA-seq) data, while for single-cell chromatin accessibility sequencing data, the epigenomic profiles at the cellular level, there is a significant gap in the availability of effective methods. Here we present Cofea, a correlation-based framework that focuses on the correlation between accessible chromatin regions, to accurately select scCAS datas features which are highly relevant to biological processes. With various simulated datasets, we quantitively demonstrate the advantages of Cofea for capturing cellular heterogeneity of imbalanced cell populations or differentiation trajectories. We further demonstrate that Cofea outperforms existing feature selection methods in facilitating downstream analysis, particularly in cell clustering, on a wide range of real scCAS datasets. Applying this method to identification of cell type-specific peaks and candidate enhancers, pathway enrichment analysis and partitioned heritability analysis, we show the potential of Cofea to uncover functional biological process and the genetic basis of cellular characteristics.