nichecompass.nn.MaskedOmicsFeatureDecoder

class nichecompass.nn.MaskedOmicsFeatureDecoder(modality, entity, n_prior_gp_input, n_addon_gp_input, n_cat_covariates_embed_input, n_output, mask, addon_mask, masked_features_idx, recon_loss)

Masked omics feature decoder class.

Takes the latent space features z (gp scores) as input, and has a masked layer to decode the parameters of the underlying omics feature distributions.

Parameters:

  • modality (Literal['rna', 'atac']) – Omics modality that is decoded. Can be either rna or atac.

  • entity (Literal['target', 'source']) – Entity that is decoded. Can be either target or source.

  • n_prior_gp_input (int) – Number of maskable prior gp input nodes to the decoder (maskable latent space dimensionality).

  • n_addon_gp_input (int) – Number of non-maskable add-on gp input nodes to the decoder ( non-maskable latent space dimensionality).

  • n_cat_covariates_embed_input (int) – Number of categorical covariates embedding input nodes to the decoder (categorical covariates embedding dimensionality).

  • n_output (int) – Number of output nodes from the decoder (number of omics features).

  • mask (Tensor) – Mask that determines which masked input nodes / prior gp latent features z can contribute to the reconstruction of which omics features.

  • addon_mask (Tensor) – Mask that determines which add-on input nodes / add-on gp latent features z can contribute to the reconstruction of which omics features.

  • masked_features_idx (List) – Index of omics features that are included in the mask.

  • recon_loss (Literal['nb']) – The loss used for omics reconstruction. If nb, uses a negative binomial loss.

Methods table

forward(z, log_library_size[, ...])

Forward pass of the masked omics feature decoder.

Methods

MaskedOmicsFeatureDecoder.forward(z, log_library_size, cat_covariates_embed=None, dynamic_mask=None)

Forward pass of the masked omics feature decoder.

Parameters:

  • z (Tensor) – Tensor containing the latent space features.

  • log_library_size (Tensor) – Tensor containing the omics feature log library size of the nodes.

  • dynamic_mask (Optional[Tensor] (default: None)) – Dynamic mask that can change in each forward pass. Is used for atac modality: if a gene is removed by regularization in the rna decoder (its weight is set to 0), the corresponding peaks will be marked as 0 in the dynamic_mask.

  • cat_covariates_embed (Optional[Tensor] (default: None)) – Tensor containing the categorical covariates embedding (all categorical covariates embeddings concatenated into one embedding).

Return type:

Tensor

Returns:

nb_means: The mean parameters of the negative binomial distribution.