Deep Learning Driven Cell-Type-Specific Embedding for Inference of Single-Cell Co-expression Networks

Abstract

The inference of gene co-expression module in specific cell types is important for understanding of cell-type-specific biological processes. We have developed DeepCSCN, an unsupervised deep-learning framework, to infer gene co-expression modules from single-cell RNA sequencing (scRNA-seq) data. Utilizing a global-to-local network construction approach, DeepCSCN can infer co-expression at the whole sample level and construct cell-type-specific co-expression networks. Systematic evaluations on eight public scRNA-seq datasets show that DeepCSCN significantly outperforms eight existing methods in co-expression network construction. Furthermore, DeepCSCN effectively identifies cell-type-specific co-expression networks that are more enriched for cell-specific functional pathways compared to current methods. Finally, application of DeepCSCN on public scRNA-seq data revealed 280 cell-specific gene modules across 27 cell types, including epithelial cells, immune cells, and myonuclei, demonstrating its versatility and accuracy in elucidating cell-type-specific gene co-expression regulation. DeepCSCN offers a powerful tool for researchers to dissect the intricate gene co-expression networks within distinct cell types.