Building gene regulatory networks from scATAC-seq and scRNA-seq using Linked Self-Organizing Maps

Abstract

Rapid advances in single-cell assays have outpaced methods for analysis of those data types.Different single-cell assays show extensive variation in sensitivity and signal to noise levels. In particular, scATAC-seq generates extremely sparse and noisy datasets. Existing methods developed to analyze this data require cells amenable to pseudo-time analysis or require datasets with drastically different cell-types. We describe a novel approach using self-organizing maps (SOM) to link scATAC-seq and scRNA-seq data that overcomes these challenges and can generate draft regulatory networks. Our SOMatic package generates chromatin and gene expression SOMs separately and combines them using a linking function. We applied SOMatic on a mouse pre-B cell differentiation time-course using controlled Ikaros over-expression to recover gene ontology enrichments, identify motifs in genomic regions showing similar singlecell profiles, and generate a gene regulatory network that both recovers known interactions and predicts new Ikaros targets during the differentiation process. The ability of linked SOMs to detect emergent properties from multiple types of highly-dimensional genomic data with very different signal properties opens new avenues for integrative analysis of single-cells.

STATegra: a comprehensive multi-omics dataset of B-cell differentiation in mouse

“…This is a transnational initiative to develop methods, software and data for dynamic multi-omics analyses. From the STATegra project several tools for integrative multi-omics data analyses have been published and released [26][27][28][29][30][31][32][33] .…”

“…STATegra uses a well-studied system, namely mouse pre-B-cell differentiation, in a cell line model 34 . This is a highly reproducible in vitro system [33][34][35][36] that allows the generation of sufficient material to deploy a comprehensive set of omics measurements. STATegra covers the three types of biomolecules and the different layers that comprise the basic flow of genetic information: chromatin structure (through DNase-seq, RRBS and ChIP-seq), gene expression (RNA-seq and miRNA-seq), proteomics and metabolomics.…”

“…ChIP-seq, DNase-seq, RNA-seq and scRNA-seq datasets were used in Vidal et al 37 to describe the cross-talk between IKAROS Foxo1 and Myc transcription factors in regulating B-cell development. scATAC-seq, scRNA-seq and ATAC-seq data have been used to develop new statistical methods for the integration of single-cell multi-omics 33 . Figure 1 illustrates the STATegra dataset.…”