SCING: Single Cell INtegrative Gene regulatory network inference elucidates robust, interpretable gene regulatory networks

Abstract

Gene regulatory network (GRN) inference is an integral part of understanding physiology and disease. Single cell/nuclei RNAseq (scRNAseq/snRNAseq) data has been used to elucidate cell-type GRNs; however, the accuracy and speed of current scRNAseq-based GRN approaches are suboptimal. Here, we present Single Cell INtegrative Gene regulatory network inference (SCING), a gradient boosting and mutual information based approach for identifying robust GRNs from scRNAseq, snRNAseq, and spatial transcriptomics data. Performance evaluation using held-out data, Perturb-seq datasets, and the mouse cell atlas combined with the DisGeNET database demonstrates the improved accuracy and biological interpretability of SCING compared to existing methods. We applied SCING to the entire mouse single cell atlas, human Alzheimers disease (AD), and mouse AD spatial transcriptomics. SCING GRNs reveal unique disease subnetwork modeling capabilities, have intrinsic capacity to correct for batch effects, retrieve disease relevant genes and pathways, and are informative on spatial specificity of disease pathogenesis.