celldisect.CellDISECT.__init__

CellDISECT.__init__(adata: anndata.AnnData, n_hidden: int = 128, n_latent_shared: int = 10, n_latent_attribute: int = 10, n_layers: int = 1, dropout_rate: float = 0.1, gene_likelihood: Literal['zinb', 'nb', 'poisson'] = 'zinb', latent_distribution: Literal['normal', 'ln'] = 'normal', split_key: str | None = None, train_split: str | List[str] = ['train'], valid_split: str | List[str] = ['valid'], test_split: str | List[str] = ['ood'], weighted_classifier=False, use_bias: bool = True, **model_kwargs)[source]

Initialize the CellDISECT model.

Parameters:

  • adata (AnnData) – AnnData object that has been registered via setup_anndata().

  • n_hidden (int, optional) – Number of nodes per hidden layer, by default 128.

  • n_latent_shared (int, optional) – Dimensionality of the shared latent space, by default 10.

  • n_latent_attribute (int, optional) – Dimensionality of the latent space for each sensitive attribute, by default 10.

  • n_layers (int, optional) – Number of hidden layers used for encoder and decoder neural networks, by default 1.

  • dropout_rate (float, optional) – Dropout rate for neural networks, by default 0.1.

  • gene_likelihood (Literal["zinb", "nb", "poisson"], optional) – Gene likelihood distribution, by default “zinb”.

  • latent_distribution (Literal["normal", "ln"], optional) – Latent distribution, by default “normal”.

  • split_key (str, optional) – Key in adata.obs to split the data into training, validation, and test sets, by default None.

  • train_split (Union[str, List[str]], optional) – Values in split_key to be used for training, by default [“train”].

  • valid_split (Union[str, List[str]], optional) – Values in split_key to be used for validation, by default [“valid”].

  • test_split (Union[str, List[str]], optional) – Values in split_key to be used for testing, by default [“ood”].

  • weighted_classifier (bool, optional) – Whether to use weighted classifiers for categorical covariates, by default False.

  • use_bias (bool, optional) – Whether to use bias terms in the neural networks, by default True.

  • **model_kwargs (dict) – Additional keyword arguments for the model.