Fig. 5: scGALA Enables Cross-modal Generation via Alignment-guided Graph Neural Networks for Extended Multi-omics Functionality. | Nature Communications

Fig. 5: scGALA Enables Cross-modal Generation via Alignment-guided Graph Neural Networks for Extended Multi-omics Functionality.

From: scGALA advances graph link prediction-based cell alignment for comprehensive data integration and harmonization

Fig. 5: scGALA Enables Cross-modal Generation via Alignment-guided Graph Neural Networks for Extended Multi-omics Functionality.

a Cross-modal RNA generation accuracy: Dot plots compare scGALA-generated RNA (from input ATAC data) to ground truth RNA, quantified by Pearson’s correlation of cell type-specific signature gene expression. b Cell type clustering fidelity: UMAPs of scGALA-generated RNA (left) and ground truth RNA (right), colored by cell type labels. Adjusted Rand Index (ARI) quantifies clustering concordance with ground truth. c Functional relevance of generated RNA: Gene Ontology Biological Process (GOBP) enrichment analysis of signature genes. Heatmaps compare (\(-{\log }_{10}(\,{\mbox{FDR}})\)) values for scGALA-generated vs. ground truth RNA, with Pearson’s correlation confirming conserved biological processes. d Cell-cell interaction preservation: Heatmap of CellChat-derived interaction strengths (Secreted Signaling and Cell-Cell Contact) for scGALA-generated RNA vs. ground truth. Pearson’s correlation validates scGALA's ability to retain multi-omics signaling dynamics. P-values are obtained from two-sided Pearson correlation tests. Source data are provided as a Source Data file.

Back to article page