Hypergraph Neural Networks Reveal Spatial Domains from Single-cell Transcriptomics Data
Soltani, M.; Rueda, L.
Show abstract
I.Spatial transcriptomics enables the measurement of gene expression while preserving spatial context within tissue samples. A key challenge is detecting spatial domains of biologically meaningful cell clusters, typically addressed using graph-based models like SpaGCN and STAGATE. However, these methods only capture pairwise relationships and fail to model complex higher-order interactions. We propose a hypergraph-based framework for spatial transcriptomics using Hypergraph Neural Networks (HGNNs). Our approach constructs hyperedges from top-K densest overlapping subgraphs and integrates histological image features and gene expression profiles. Combined with autoencoders, our model effectively learns expressive node embeddings in an unsupervised setting. Evaluated on a mouse brain dataset, our model achieves the highest iLISI score of 1.843 and outperforms state-of-the-art baselines with an ARI of 0.51 and a Leiden score of 0.60.
Matching journals
The top 4 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Predicting Spatially Resolved Gene Expression via Tissue Morphology using Adaptive Spatial GNNs 97%
- Graspot: A graph attention network for spatial transcriptomics data integration with optimal transport 96%
- GraphCompass: Spatial metrics for differential analyses of cell organization across conditions 95%
Similar papers in this journal
- Graph Contrastive Learning of Subcellular-resolution Spatial Transcriptomics Improves Cell Type Annotation and Reveals Critical Molecular Pathways 96%
- Evaluating discrepancies in dimensionality reduction for time-series single-cell RNA-sequencing data 96%
- BayeSMART: Bayesian Clustering of Multi-sample Spatially Resolved Transcriptomics Data 95%
Similar papers in this journal
Similar papers in this journal
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.