omicverse.pp.sude

omicverse.pp.sude(adata, n_pcs=None, *, no_dims=2, use_rep=None, k1=20, normalize=True, large=False, initialize='le', agg_coef=1.2, T_epoch=50, key_added=None, copy=False)[source]

SUDE (Scalable Unsupervised Dimensionality reduction via Embedding) dimensionality reduction.

Perform SUDE dimensionality reduction for visualization of single-cell data. SUDE is a scalable unsupervised dimensionality reduction method that can handle large-scale datasets efficiently by using landmark sampling and constrained locally linear embedding.

SUDE was proposed for scalable dimensionality reduction of single-cell data. It uses a two-stage approach: first computing embeddings for landmark points, then interpolating the remaining points using constrained locally linear embedding.

Parameters:

  • adata (AnnData) – Annotated data matrix.

  • n_pcs (int | None (default: None)) – Number of principal components to use for preprocessing. If None, uses the full data matrix.

  • use_rep (str | None (default: None)) – Key for the representation to use. If None, uses adata.X.

  • no_dims (int (default: 2)) – The number of dimensions of the embedding.

  • k1 (int (default: 20)) – Number of nearest neighbors for PPS (Probabilistic Point Sampling) to sample landmarks. Must be smaller than the number of observations. If set to 0, all points are used as landmarks.

  • normalize (bool (default: True)) – Whether to normalize the data using min-max normalization. Should be set to True if features are on different scales.

  • large (bool (default: False)) – Whether to use the large-scale version that splits data into blocks to avoid memory overflow. Recommended for datasets with >10k cells.

  • initialize (str (default: 'le')) – Method for initializing the embedding before manifold learning. Options: ‘le’ (Laplacian eigenmaps), ‘pca’ (PCA), ‘mds’ (MDS).

  • agg_coef (float (default: 1.2)) – Aggregation coefficient for computing modified distance matrix. Controls the influence of shared nearest neighbors.

  • T_epoch (int (default: 50)) – Maximum number of epochs for optimization.

  • key_added (str | None (default: None)) – If not specified, the embedding is stored as obsm[‘X_sude’] and the parameters in uns[‘sude’]. If specified, the embedding is stored as obsm[key_added] and the parameters in uns[key_added].

  • copy (bool (default: False)) – Return a copy instead of writing to adata.

Return type:

AnnData | None

Returns:

  • Returns None if copy=False, else returns an AnnData object. Sets the following fields

  • `adata.obsm[‘X_sude’ | key_added]` (numpy.ndarray (dtype float)) – SUDE coordinates of data.

  • `adata.uns[‘sude’ | key_added]` (dict) – SUDE parameters.

Examples

>>> import omicverse as ov
>>> adata = ov.datasets.pbmc3k()
>>> ov.pp.sude(adata)
>>> ov.pl.sude(adata, color='leiden')