Detecting Coupled-Gene Clusters In scRNA-Seq Data Using Deep Learning

Alicia Petrany, Yong Chen PhD.

The characterization of gene co-expression and co-differential expression across single cells can provide biological insights into pathway mapping and gene expression network development. Single-cell RNA sequencing (scRNA-Seq) has been widely used to investigate diverse biological functions due to its unprecedented high resolution, however it remains open to detect coupled-gene clusters whose genes have synchronized transcriptional dynamics across multiple experimental conditions. Here we present a deep-learning-based method, DECUPLE, to characterize coupled-gene clusters in scRNA-seq datasets. DECUPLE determines gene profile similarity by computing Pearson similarity metrics on permutated and pooled gene-wise expression levels of multiple cellular conditions. Then, DECUPLE applies network node embeddings through self-supervised learning on multiplex networks with high-order mutual information. Finally, DECUPLE clusters the node embeddings to ultimately characterize coupled gene expression. We validated DECUPLE on real scRNA-Seq datasets and demonstrated its ability to identify functionally distinct gene clusters.