Gene regulatory network inference by reducing dimensions for single-cell transcriptome profiles

Abstract

Understanding the complex regulatory relationships between genes is central to conducting cellular functionality and disease. However, the inference of Gene Regulatory Networks (GRNs) from sparse single-cell RNA sequencing (scRNA-seq) data is difficult from a computational perspective, especially when nonlinear dependence is a common feature. To address this, we developed NIRD: Network Inference by Reduced Dimension, a new Linux command line tool for large-scale GRN inference. NIRD relies on applying matrix-factoring strategies to obtain a more manageable representation of scRNA-seq profiles that still preserves critical biological signals. Each matrix-factoring strategy was utilized to target linear versus complex, non-linear relationships that standard linear methods may not capture, and performance for each matrix-factoring strategy was evaluated for GRN inference using the Area Under the Curve (AUC) score. NIRD outputs a complete gene-by-gene matrix representative of inferred regulatory interactions for the cellular population. We expect it will provide an insightful and efficient process of ultimately extracting more informative knowledge from complex regulatory interactions within single-cell transcriptome data.