netmap.grn.inferrence.attribution_one_target

netmap.grn.inferrence.attribution_one_target(target_gene, lrp_model, input_data, xai_type='lrp-like', background_type='zeros')[source]

Compute Captum attributions for a single target gene across all models.

Iterates over every explainer in lrp_model and calls .attribute() with the appropriate signature depending on xai_type. The background tensor used for SHAP-style attribution is constructed from input_data according to background_type.

Parameters:

  • target_gene (int) – Index of the target gene in the expression matrix.

  • lrp_model (list) – List of Captum explainer objects (one per model in zoo).

  • input_data (torch.Tensor) – Input data tensor on CUDA, shape (n_cells, n_genes).

  • xai_type (str) – 'lrp-like' or 'shap-like'.

  • background_type (str) – Background for SHAP-like methods — 'zeros' (default), 'randomize' (shuffle columns), or 'data' (use input as its own baseline).

Returns:

One attribution matrix per model, each of shape

(n_cells, n_genes).

Return type:

list