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

摘要

Gene expression is a tightly controlled process occurring in all cells during all stages of organismal life. How much and when genes are expressed is determined by gene regulatory networks (GRNs), which encode the biological programs that cells can perform. Each cell in an organism is constantly running through these networks to carry out its particular function. New techniques allow us to measure gene expression and chromatin accessibility using single-cell RNA-seq (scRNA-seq) and single-cell ATAC-seq (scATAC-seq). However, these techniques have relatively poor and different signal-to-noise ratios. In this work, we use a form of unsupervised learning called Self-Organizing Maps (SOMs) to analyze one step of B cell differentiation by linking separately trained scRNA-seq and scATAC-seq SOMs. We mine the linked SOMs to reconstruct the underlying GRN using a top-down approach. The resulting GRN not only recapitulated known regulatory linkages but also identified a large number of potential regulatory connections to the system. These methods should be generally applicable to linking heterogeneous high-throughput data with different signal-to-noise profiles.