An Exploration of Integrative Analysis of Simultaneously Profiled scHi-C and scRNA-Seq Data Using Graph Convolutional Network

Understanding the relationship between three-dimensional genome organization and gene expression is key to unraveling cellular identity. In this study, we explore an integrative analysis method for simultaneously profiled scHi-C and scRNA-seq data using unsupervised Graph Convolutional Networks (GCNs). By representing chromatin interactions and transcriptional activity as graphs, we analyze single-cell data, with nodes capturing genomic bins and edges reflecting chromatin interactions. Using unsupervised GCNs, we first train models on each dataset, then extract embeddings, reduce their dimensionality with UMAP, and cluster them using k-means. We applied our framework to simultaneously profiled datasets of diverse mouse cell types. Performance metrics included Adjusted Rand Index (ARI), Silhouette scores, and UMAP visualizations. Our key findings are: (1) While scHi-C alone highlighted genome structures (Silhouette median ~ 0.6), integrating scRNA-seq did not consistently enhance clustering. (2) Low ARI scores (median ~ 0) indicate discrepancies between clustering and biological annotations, suggesting the complexity of uncovering novel genomic insights. Further refinement is needed to align computational outputs with biological interpretations.