troutpy.tl.latent_factor

Contents

troutpy.tl.latent_factor#

troutpy.tl.latent_factor(sdata, method='NMF', layer='segmentation_free_table', n_components=20, copy=None, random_state=None, drvi_model_path=None, **kwargs)#

Apply latent factor identification (NMF, LDA, or DRVI) to reduce the dimensionality of gene expression data.

Parameters:

  • sdata (spatialdata.SpatialData) – SpatialData object with the specified layer containing AnnData.

  • method (str, optional) – One of "NMF", "LDA", or "DRVI". Defaults to "NMF".

  • layer (str, optional) – The AnnData layer in sdata to operate on. Defaults to "segmentation_free_table".

  • n_components (int, optional) – Number of latent dimensions (ignored if a pretrained DRVI model is loaded). Defaults to 20.

  • copy (bool, optional) – If truthy, return the modified SpatialData object; otherwise modify sdata in place and return None.

  • random_state (int, optional) – Random seed for "NMF" and "LDA".

  • drvi_model_path (str, optional) – Path to a pretrained DRVI model to load instead of training a new one.

  • **kwargs – Additional parameters. For "NMF"/"LDA" these are forwarded to model.fit_transform/model.fit. For "DRVI" the following are popped before constructing/training the model: encoder_dims (list of int), decoder_dims (list of int), n_epochs (int), kl_warmup (int), is_count_data (bool), early_stopping (bool), accelerator (str), devices (int).

Returns:

spatialdata.SpatialData or None Modified SpatialData object if copy is truthy; otherwise None.

Contents