omicverse.space.Cal_Spatial_Net

omicverse.space.Cal_Spatial_Net(adata, rad_cutoff=None, k_cutoff=None, max_neigh=50, model='Radius', verbose=True)[source]

Construct spatial neighbor networks for spatial integration.

This function builds a spatial neighborhood graph by connecting spots based on their physical distances. It supports both radius-based and k-nearest neighbor approaches for network construction.

Parameters:

  • adata (AnnData) – Spatial AnnData containing coordinates in obsm['spatial'].

  • rad_cutoff (float, optional) – Radius threshold when model='Radius'.

  • k_cutoff (int, optional) – Number of nearest neighbors when model='KNN'.

  • max_neigh (int, default=50) – Maximum neighbors queried before filtering by model.

  • model ({'Radius', 'KNN'}, default='Radius') – Strategy for building spatial edges.

  • verbose (bool, default=True) – Whether to print graph statistics.

Returns:

  • None – Writes adata.uns['Spatial_Net'] and adata.uns['adj'].

  • Notes –

    • For STAligner, adjust rad_cutoff to ensure 5-10 neighbors per spot

    • Includes self-loops in adjacency matrix

    • Uses ball_tree algorithm for efficient neighbor search

    • Memory efficient implementation for large datasets

    • Critical for downstream integration tasks