Welcome to GRASP’s page!¶
What’s GRASP for?¶
GRASP is a geometric deep learning framework designed to discover and quantify subcellular mRNA localization patterns from spatial transcriptomics data, capturing cell-to-cell heterogeneity without predefined categories or assumptions.
What’s the model?¶
The GRASP model combines a Graph Attention Network (GAT) for representation learning with a momentum contrastive learning module. It builds rotation-invariant transcript spot graphs (TSGs), learns embeddings through intra- and inter-cellular positive pairs, and optimizes both contrastive and spatial reconstruction objectives for robust pattern representation.