scSAGRN: Inferring gene regulatory networks from single-cell multi-omics using spatial association

Abstract

Identifying the regulatory relationships between transcription factors and target genes is fundamental to understanding molecular regulatory mechanisms in biological processes including development and disease occurrence. Therefore, resolving the relationships between cis-regulatory elements and genes using single-cell multi-omics data is important for understanding transcriptional regulation. Here, scSAGRN is proposed as a framework for inferring gene regulatory networks from single-cell multi-omics. scSAGRN incorporates spatial association to compute correlations between gene expression and chromatin openness data, connects distal cis-regulatory elements to genes, infers gene regulatory networks and identifies key transcription factors. The approach is benchmarked using real single-cell datasets, and scSAGRN shows superior performance in TF recovery, peak-gene linkage prediction, and TF-gene linkage prediction compared to existing methods. Meanwhile, in human peripheral blood mononuclear cells dataset, mouse cerebral cortex dataset and mouse embryonic brain cells dataset, scSAGRN demonstrates its capability to infer gene regulatory networks and identify transcription factors. Overall, scSAGRN provides a reference for predicting transcriptional regulatory patterns from single-cell multi-omics data.