Multi-Network Co-expression Analysis Enhances Biological Insights from Single-Cell Gene Expression.

With the advent of single-cell and single-nucleus RNA sequencing (sc/snRNA-seq), we can build cell-type-specific gene co-expression networks (GCNs). However, the high sparsity of scRNA-seq data limits GCN construction. We present scCoExpNets, a framework to create and annotate single-cell GCNs. For each cell-type cluster, scCoExpNets generates an initial GCN (T0) from its expression matrix. To reduce sparsity, pseudo-cells are created and used to build an alternative GCN (T1). This process is repeated through several iterations, thus creating multiple matrices and multiple GCNs for the same cell type. scCoExpNets was applied to an snRNA-seq dataset from the substantia nigra pars compacta (SNpc) from post-mortem samples of 13 controls and 14 Parkinson’s disease (PD) cases. Thanks to the creation of multiple GCNs for each cell type cluster, scCoExpNets detected that 95.84% of the T0 modules change their gene composition after sparsity reduction. In consequence, new biological annotations emerge, including 183 Ti GO subgraphs not detected at previous iterations, 159 GO subgraphs that expanded, and 157 GO subgraphs that specialized T0 GO subgraphs. Finally, we showcase the capabilities of scCoExpNets to detect a well-maintained dopaminergic module enriched with ferroptosis suppressors (Padj<0.004), which was replicated in two independent SNpc datasets (3.39 and 12.17 Z-summary score). The scCoExpNets R package and the GCNs created for all cell types are available on GitHub.