Dual-graph topology alignment for single-cell multi-omics integration

Single-cell multi-omics technologies provide unprecedented opportunities to characterize cellular heterogeneity and regulatory programs by jointly profiling complementary molecular layers. However, effective integration of heterogeneous modalities remains challenging because different omics data reside in distinct feature spaces and often contain noisy cross-modality correspondences. Here, we present Dual-Graph Topology Alignment (DGTA), a reference-guided graph-based framework for single-cell multi-omics integration that combines hybrid graph filtration with dual-graph topology alignment. DGTA preserves within-modality topology, refines noisy cross-modality correspondences through label-guided graph filtration, and jointly optimizes supervised classification, cross-modality consistency, and query-graph regularization to learn a shared latent representation. Across multiple benchmark datasets, including human myocardial infarction data, mouse atlas data, and multimodal PBMC data, DGTA showed competitive performance in label transfer accuracy, clustering consistency, and biological structure preservation. In addition, DGTA successfully aligned transcriptomic, chromatin accessibility, and DNA methylation profiles within a shared latent representation, demonstrating its extensibility to more complex multi-omics settings. These results indicate that DGTA provides an effective and extensible framework for topology-aware single-cell multi-omics integration.