omicverse.space.pySpaceFlow

omicverse.space.pySpaceFlow#

class omicverse.space.pySpaceFlow(adata)[source]#

SpaceFlow spatial flow analysis class.

SpaceFlow is a deep learning method for analyzing spatial transcriptomics data by learning spatially-aware cell representations. It combines graph neural networks with spatial regularization to capture both transcriptional and spatial relationships between cells.

The method: 1. Constructs a spatial neighborhood graph 2. Learns embeddings using deep graph infomax 3. Applies spatial regularization to preserve spatial structure 4. Generates pseudo-spatial maps for trajectory analysis

Parameters:

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

  • Attributes –

    adata: AnnData

    Input annotated data matrix containing: - Gene expression data in adata.X - Spatial coordinates in adata.obsm[‘spatial’]

    sf: SpaceFlow

    Internal SpaceFlow object for computations

    embedding: array

    Learned spatial-aware embeddings after training

  • Examples – 

    >>> import scanpy as sc
>>> import omicverse as ov
>>> # Load spatial data
>>> adata = sc.read_visium(...)
>>> # Initialize SpaceFlow
>>> spaceflow = ov.space.pySpaceFlow(adata)
>>> # Train model
>>> embedding = spaceflow.train(
...     spatial_regularization_strength=0.1,
...     z_dim=50,
...     epochs=1000
... )
>>> # Calculate pseudo-spatial map
>>> psm = spaceflow.cal_pSM(n_neighbors=20)

__init__(adata)[source]#

Initialize SpaceFlow spatial analysis object.

Parameters:

  • adata (AnnData) – Spatial AnnData used for SpaceFlow embedding.

  • Notes –

    • Automatically checks for SpaceFlow package installation

    • Constructs initial spatial neighborhood graph

    • Uses 10 nearest neighbors for graph construction

    • Stores SpaceFlow object in self.sf

Methods

__init__(adata)

Initialize SpaceFlow spatial analysis object.

cal_pSM([n_neighbors, resolution, ...])

Calculate pseudo-spatial map using diffusion pseudotime.

train([spatial_regularization_strength, ...])

Train SpaceFlow model for spatial embedding.
Contents