knn_normalization.tl.knn_normalize

Contents

knn_normalization.tl.knn_normalize#

knn_normalization.tl.knn_normalize(data, calculate_neighbors_from='prot', preprocess_rna=True, log_transform=True, n_neighbors=None, pseudocount=5, max_iterations=25, mean='average', inplace=True, save_size_factors=False, verbose=True, preserve_total_counts=True)#

Normalize protein expression with KNN normalization.

Parameters:

  • data (AnnData | MuData) – AnnData object with protein expression counts or MuData object with prot and rna modalities.

  • calculate_neighbors_from (Literal['prot', 'rna', 'use_existing_neighbors'] (default: 'prot')) – Whether to use the prot or the rna modality to calculate neighbor cells. If use_existing_neighbors, the neighbors already present in the protein data will be used.

  • preprocess_rna (bool (default: True)) – If using RNA to calculate neighbors, whether to preprocess the RNA data with library size normalization and log-transformation first.

  • n_neighbors (Optional[Integral] (default: None)) – Number of neighbors. If None, calculated automatically as max(15, min(round(n_cells / 20), 300)).

  • log_transform (bool (default: True)) – Whether to log-transform the protein data.

  • pseudocount (Integral (default: 5)) – Pseudocount to add before normalization to avoid zero-division errors.

  • max_iterations (Integral (default: 25)) – Maximum number of iterations.

  • mean (Literal['average', 'geom_mean', 'trimmed_mean'] (default: 'average')) – Type of mean to use: 'average', 'geom_mean', or 'trimmed_mean'.

  • inplace (bool (default: True)) – Whether to update the object in place or return a copy.

  • save_size_factors (bool (default: False)) – If True, saves the final size factors to data.obs['size_factor'] and the size factor history to data.obsm['size_factor_history'].

  • verbose (bool (default: True)) – Whether to print progress messages.

  • preserve_total_counts (default: True) – Whether to preserve total counts across iterations.

Returns:

Normalized data will be written to data (if it is an AnnData object) or data.mod['prot'] (if it is a MuData object) as an X matrix. If inplace is False, returns a new AnnData object (if input is AnnData) or a new MuData object (if input is MuData) with the normalized data.

Contents