Multi-omics single cell data integration and clustering based on attribute graph embedding

Integrating multi-omics data in single-cell analyses is critical to improving our understanding of cellular heterogeneity and regulatory mechanisms. To address the shortcomings in capturing the complexity of cell states and interactions, we propose an innovative integrated framework based on property graph embedding, utilizing the Adaptive Graph encoder (AGE) model. Our approach efficiently embeds high-dimensional omics features and complex graph structures to generate robust low-dimensional cell representations. Through the joint analysis of transcriptomic and epigenetic data, the model can realize single-cell multi-omics integration and cell type clustering, which is significantly superior to traditional matrix decomposition, graph network and transfer learning methods. Through the verification of different real data sets, the experimental results show that compared with other traditional methods, the clustering accuracy is improved by about 55%, and the robustness is enhanced. This study provides a universal tool for cell analysis and promotes in-depth biological research.