RMGAT: a Relational Multi-Scale Graph Attention Network for spatial transcriptomics domain identification with systematic component ablation
A systematic ablation study providing evidence-based design guidance for spatial transcriptomics GNNs is provided, with findings support using GAT attention and contrastive learning and investing in node feature quality over edge feature complexity.