Title: HyperST: a unified framework for integrating intracellular gene interaction and cell spatial interaction
Problem or question being addressed
The physiological functions of multicellular tissues are not only defined by heterogeneous cells forming these tissues but are also highly dependent on complicated local and distal cell-cell interactions [1, 2]. Recent advances in various spatially resolved transcriptome profiling techniques have made it possible to measure gene expression profiles at single-cell or subcellular resolution while simultaneously retaining information on the spatial locations of cells [3]. How to model intracellular and extracellular interactions in space in a systematic way, and then mine the hidden correlations between intracellular molecular processes and cellular localization from such complex data, for discovering gene regulatory pathways that determine cell interactions, identifying intracellular processes that are affected by cell interactions, have not been adequately answered. To investigate these questions, an analytical framework that integrates intracellular gene regulation and cellular interactions is required.
Rationale and details of suggested approach
For the design of HyperST, I borrowed a concept called hypergraph to convert single-cell spatial transcriptome data into a hypergraph (Fig. 1). The difference between a hypergraph and a normal graph is that hyperedges in hypergraph could connect more than two nodes. In the hypergraph, there are four kinds of nodes, ligand genes, receptor genes, cell node, and other intracellular gene nodes. In addition, there are two kinds of hyperedges, one is a hyperedge that connects a cell and a gene pair, indicating that the interaction of gene pair is activated in the cell. The other is a cell pair and a pair of ligand genes, suggesting the cell pair interacts through the pair of LR genes. This hypergraph also includes the spatial neighboring relationships of cells. With this design, various complex relationships inside and outside cells could be converted into the problem of hypergraph embedding, and the potential for extracellular interactions not observed by experimental results could be changed into the problem of hyperedge prediction (Fig. 1). The core idea is to use the known LR gene-induced cell-to-cell interactions, combined with spatial localization and intracellular gene regulatory relationships, to infer other gene relationships that lead to cell-to-cell interactions.
Fig. 1 Overview of HyperST. There are four kinds of nodes and two kinds of hyperedges in the original hypergraph.
How it will affect the broader field
Integrating the intracellular and intercellular interactions, the advantages of spatial omics technology could be fully exploited, and the bidirectional correlations between multi-dimensional molecular signatures and cell assembly in tissues could be maximally mined. In our method, cell interactions caused by known ligand-receptor pairs are the object of learning, so that a broader range of potential interaction mechanisms (key molecular features, gene combinations and clusters, pathways, etc.) can be directly or indirectly inferred. Our preliminary results have shown that HyperST can effectively infer the molecular mechanisms on which long-range interactions depend (not limited to ligands, receptors and downstream genes), such as ROS mechanisms, which has never been achieved in previous analyses.
References
1 Gunzer, M. Migration, cell-cell interaction and adhesion in the immune system. Ernst Schering Foundation Symposium Proceedings 2, 97 (2007).
2 Armingol, E., Officer, A., Harismendy, O. & Lewis, N. E. Deciphering cell-cell interactions and communication from gene expression. Nat Rev Genet 22, 71-88, doi:10.1038/s41576-020-00292-x (2021).
3 Liao, J., Lu, X., Shao, X., Zhu, L. & Fan, X. Uncovering an Organ's Molecular Architecture at Single-Cell Resolution by Spatially Resolved Transcriptomics. Trends Biotechnol 39, 43-58, doi:10.1016/j.tibtech.2020.05.006 (2021).
4 Zhou D , Huang J , Schlkopf B . Learning with Hypergraphs: Clustering, Classification, and Embedding, Proceedings of the Twentieth Annual Conference on Neural Information Processing Systems, 2006.