scHG: a supercell framework with high-order graph learning enables hyper-fast multi-omics analysis

Abstract

Multi-omics profiling--spanning proteomics, transcriptomics, and additional omics data types--is rapidly advancing, providing increasingly detailed maps of cellular identity and function. In parallel, computational methods have evolved to integrate diverse omics with increasing precision in resolving cellular heterogeneity and delineating subtle subpopulation structures. Yet, capturing rare cell population while maintaining computational tractability remains a major challenge. Here, we introduce the supercell paradigm, in which expression-coherent cells are compressed into candidate units for rare cell population. Supercells are constructed using angle-aware similarity metrics and second-order co-occurrence neighbors, with impurity cells pruned by degree centrality. To address scalability, we implement sparse matrix optimization and iterative high-order graph updates, enabling efficient integration of large-scale multi-omics datasets. Building on this framework, we develop scHG, a high-order graph learning approach guided by an omics-weighted optimizer that adaptively balances contributions from gene expression, surface proteins, and chromatin accessibility. Across six benchmark datasets (up to 30672 cells), scHG consistently outperforms state-of-the-art methods, improving mean ARI and NMI by 3.97% and 3.54%, respectively, while reducing runtime by 26.40%. Beyond performance gains, scHG was able to resolve fine-grained cellular heterogeneity within conventionally defined T cell populations, distinguishing subpopulations such as memory-like or progenitor-exhausted T cells and innate-like cytotoxic T cells. At the same time, the supercell framework uncovered rare populations, including dendritic-cell populations and NK-like B cells, that remained masked at the cluster level under standard pipelines. These results underscore both the rare-cell detection capability and the computational efficiency of scHG. Our code and data are available at http://mialab.ruc.edu.cn/scHG_code/zip.