from cellucid import show_anndata # In-memory AnnData show_anndata(adata) # From h5ad file (lazy loading) show_anndata("/path/to/data.h5ad") # From zarr store (lazy loading) show_anndata("/path/to/data.zarr")

from cellucid import prepare, show # Pick UMAP embeddings (explicit keys preferred, fall back to X_umap). X_umap_1d = adata.obsm.get("X_umap_1d") X_umap_2d = adata.obsm.get("X_umap_2d") X_umap_3d = adata.obsm.get("X_umap_3d") X_umap = adata.obsm.get("X_umap") # may be 1D/2D/3D depending on how it was computed if X_umap is not None and X_umap_1d is None and X_umap.shape[1] == 1: X_umap_1d = X_umap if X_umap is not None and X_umap_2d is None and X_umap.shape[1] == 2: X_umap_2d = X_umap if X_umap is not None and X_umap_3d is None and X_umap.shape[1] == 3: X_umap_3d = X_umap # Optional: include any per-cell displacement vector fields in embedding space # (see "Vector Field Overlay" section below for naming conventions). vector_fields = { "velocity_umap_2d": adata.obsm.get("velocity_umap_2d"), "velocity_umap_3d": adata.obsm.get("velocity_umap_3d"), } prepare( # Required inputs latent_space=adata.obsm.get( "X_pca", X_umap_3d if X_umap_3d is not None else (X_umap_2d if X_umap_2d is not None else X_umap_1d), ), obs=adata.obs, var=adata.var, gene_expression=adata.X, # Optional inputs connectivities=adata.obsp.get("connectivities"), vector_fields=vector_fields, # Embeddings (provide any/all of 1D/2D/3D) X_umap_1d=X_umap_1d, X_umap_2d=X_umap_2d, X_umap_3d=X_umap_3d, # Output + performance knobs out_dir="./my_export", compression=6, var_quantization=8, obs_continuous_quantization=8, ) # View anytime (fastest loading) show("./my_export")

# Required: one of these adata.obsm['X_umap_3d'] # shape (n_cells, 3) - recommended adata.obsm['X_umap_2d'] # shape (n_cells, 2) adata.obsm['X_umap'] # shape (n_cells, 2 or 3) # Optional adata.obsm['velocity_umap_2d'] # shape (n_cells, 2) - vector field overlay (velocity/drift) adata.obs # Cell metadata (categorical/continuous) adata.X # Gene expression (dense or sparse) adata.obsp['connectivities'] # KNN graph edges

from cellucid import add_transition_drift_to_obsm add_transition_drift_to_obsm( adata, T_fwd, # (n_cells, n_cells) sparse/dense transition matrix basis="umap", field_prefix="T_fwd", # -> writes adata.obsm['T_fwd_umap_<dim>d'] )

import scanpy as sc sc.pp.neighbors(adata) sc.tl.umap(adata, n_components=3) adata.obsm['X_umap_3d'] = adata.obsm['X_umap']

from cellucid import show_anndata viewer = show_anndata(adata, height=600) # Programmatic control viewer.highlight_cells([1, 2, 3], color="#ff0000") viewer.set_color_by("cell_type") viewer.set_visibility([0, 1, 2], visible=False) # Cleanup viewer.stop()