spared.layer_operations.add_noisy_layer

spared.layer_operations.add_noisy_layer(adata: AnnData, prediction_layer: str) → AnnData

Add an artificial noisy layer.

This function should only be used for experimentation/ablation purposes. The noisy layer is created by returning the missing values to an already denoised layer of the adata. In the case the 'prediction_layer' is on \(\log_2(TPM+1)\) logarithmic scale, the noisy layer is created by assigning zero values to the missing values (adds 'noisy' layer to the adata). In the case the 'prediction_layer' is on delta scale, the noisy layer is created by assigning the negative mean of the gene to the missing values (adds 'noisy_d' layer to the adata). Missing values are specified by the binary adata.layers['mask'] layer that must be already present and has True values for all real data and False values for imputed data.

Parameters:

• adata (ad.AnnData) – The AnnData object to update. Must have the adata.layers[prediction_layer], the gene means if its a delta layer, and adata.layers['mask'].

• prediction_layer (str) – The layer that will be corrupted to create the noisy layer.

Returns:

The updated AnnData object with the adata.layers['noisy'] or adata.layers['noisy_d'] layer added depending on prediction_layer.

Return type:

ad.AnnData